Comprehensive review of computational intelligence based smart city community

Abstract

The density of population in cities is growing at a faster rate to make the life of people in cities comfortable and save. The city needs to be smart. It can be mainly achieved by intelligent decision making process using computational intelligence based systems. Keeping this in view, many researchers and organizations are working to develop and implement computational intelligence decision support systems. To obtain a comprehensive overview on the current status on SI based smart city community the present investigation has been made. To achieve this objective recently published standard articles on this important sub area have been collected and reviewed. The summary of the review has been presented in systematic manner to facilitate the researchers who are currently working in the area of smart city community. The important findings of the review have been made and presented. The important performance measures in various aspects of smart city obtained by the computational intelligence methods have been listed. It is expected that the findings and the contribution of the paper will benefit the researchers, the related government and private organizations in terms of furthering their research efforts and producing different smart products pertaining to community development and improvement of comfort level of the dwellers of the smart city.

Keywords

Smart city computational intelligence machine learning intelligent transportation system smart healthcare smart environment smart education cyber security

1 Introduction

It is reported [1] that the urbanization will grow from the present 54%to 66%by 2050. Further, the cities occupy only 3%of the land area of the earth but its green house gas emission is 60–80%. This will lead to severe impact on the environment, security and management of the cities. To control and manage the exponential growth of urbanization, the concept of smart cities has emerged. In many countries of the globe have adopted this so that the resources can be managed and energy consumption can be optimal. The smart city projects aim for maintaining green environment, low carbon emission and reduction of global warming. To meet the requirement of smart city, effective utilization of information, communication and artificial intelligence (AI) techniques are essential [2].

To have an in depth understanding of the role of computational intelligence (CI) methods in smart city community development, relevant recently published articles have been collected and thoroughly reviewed in this paper. The outcome of the review is systematically presented in this article so that the interested researchers working in this field as well as the associated scientists and engineers in smart city projects will be immensely benefitted.

The CI is a broad area of AI which constitutes three groups. The first group comprises of artificial neural network (ANN) and its variations. The second group deals with fuzzy logic and its variations. The biologically inspired/ nature inspired techniques constitute the evolutionary computing. The various types of deep learning (DL) come under the broad area of ANN. In this review paper, smart city community related papers employing CI technique have been chosen. The analysis of the review on this emerging area is presented in tabular forms to facilitate the new researchers.

2 Research gap, motivation of research, research objectives and organization

2.1 Research gap

The review articles [1 –6] on the theme of the present paper reveal that CI techniques have been extensively used for the smart city community. Some of the important aspects of smart community which need further investigation and improvement are: (i) Review articles on smart city (RA) (ii) intelligent transportation system (ITS) (iii) cyber security (CS) (iv) efficient utilization of smart grid (SG) (v) unmanned aerial vehicle (UAV) (vi) smart environment (SE) (vii) Smart education (SEd) (viii) swarm intelligence based optimization (SI) and (ix) smart health care (SH).

2.2 Motivation of research

For the benefit of the community living in smart cities the various developmental areas listed in Section 2.1 have to be focused keeping this in view relevant articles have been collected and analysis has been carried out. These issues are required to be addressed with suitable methodologies and implementation to achieve the objectives and benefits of smart cities.

2.3 Research objectives

Based on the motivation of the research the objectives of the proposed article are: (i) articles related to each of the six aspects stated in Section 2.1 have been collected and categorized (ii) the standard databases used in various articles have been identified and listed (iii) for each category the CI techniques used have been identified and listed (iv) the performance achieved by use of each CI technique for different aspects of smart city has been presented in tabular form (v) the contribution of the paper and the possible extensions are also presented.

2.4 Organization

The organization of the paper proceeds as follows: In Section 3, the detailed review of the articles on CI based smart city community is presented. The sources of data and the CI techniques used to solve various aspects of the smart city have been identified and explained in Section 4. In Section 5, the various CI methods employed for smart city community are dealt in brief. In Section6, the analysis of the results of various articles reviewed in this paper has been dealt. Finally, in Section 7, the major contribution of the paper and possible extensions are presented.

3 Literature review

This section presents a comprehensive view of collected articles on smart city community. In total 28 articles published in recent years in standard journals and dealing with on the theme of the paper has been collected. The summary of the review is presented in Table 1. Each row of this table is subdivided into five columns. These are (i) Title of the paper, Name of the Journal and Name of the Publisher (ii) Datasets used (iii) Problem statement and methods employed (iv) contribution of the paper and (v) conclusion.

Table 1
Summary of literature review of articles

Title of the paper, Names of the Journal and Publisher Dataset used Problem statement and methods used Contribution of the paper Conclusion

1. Smart energy systems for sustainable smart cities: current developments, trends and future directions.Elsevier Applied Energy –Individual paper of this review article employs its own dataset. –The articles related to smart energy systems using CI in smart cities have been reviewed. –The paper has outlined current developments, the present status and further research scopes. –The present trend of energy management system using CI has been summarized and presented. It also provides the future direction of energy management in smart cities.

2. Applications of artificial intelligence and machine learning in smart cities. Elsevier Computer Communications –Individual paper of this review article employs its own dataset. –The paper has reviewed CI based articles related to energy, transportation, communication, healthcares related to smart city. –The paper has outlined the present scenario of smart cities in terms of power, communication, healthcare and transport. –The paper has dealt with shortcomings of smart cities and scope of future development using CI based methods.

3. A survey on an emerging area: Deep learning for smart city data EEE IEEE Transactions on emerging topics in computational intelligence. –Individual paper of this review article employs its own dataset. –The papers related to the problems of the smart city using deep learning have been surveyed and the findings are presented. –It is reported that there are many challenges in smart city which needs solution using deep learning. These have been dealt in the paper in addition to existing solutions –The security and knowledge fusion issues need to be addressed as future extension of work.

4. A review on deep learning for future smart cities. Wiley Internet Technology Letters –Individual paper of this review article employs its own dataset. –A review papers on future smart cities addressing on traffic mobility of citizen and drainage through IoT and using deep learning approach is presented. –The paper outlines the various deep learning based approaches to solve smart city problems. –The DL has been applied to the field of transportation, infrastructure, governance, education, healthcare and security.–The challenges are also outlined.

5. Big data analysis in intelligent transportation systems: A surveyIEEE IEEE Transactions on intelligent transportation systems –Individual paper of this review article employs its own dataset. –A survey paper on intelligent transportation systems using big data analytics is presented. –The big data analytics on ITS have provided solution relating to road traffic accident, traffic flow prediction and rail transport management. –Feasible solutions of transportation systems using intelligent data analytics approaches have been provided. It also outlines open challenges related to big data analytics.

6. Artificial intelligence for smart renewable energy sector in Europe- Smart Energy Infrastructures for next generation smart cities. IEEE IEEE Access –Individual paper of this review article employs its own dataset. –The paper provides a summary of review articles on smart energy infrastructure in smart cities of Europe employing CI techniques. –The paper has analyzed issues related to energy consumption labor productivity and implication of using CI for smart cities. –The paper outlines the development of a framework in energy sector using CI technique. It also discusses the implications for smart city research and future research scopes.

7. LSTM network: a deep learning approach for short-term traffic forecast. IET IET Intelligent transport systems Traffic data collected from Beijing traffic management bureau Forecasting of short term traffic.–A deep LSTM network is used. –The proposed LSTM model achieves better short term forecasting of traffic compared to ARIMA, SVM, RBF, SAE and RNN method. –The paper has developed LSTM based traffic volume prediction model. The work can be extended to build multiple output traffic forecast system.

8. Big data applications in real-time traffic operation and safety monitoring and improvement on urban expressways. Elsevier Transportation Research Part C In 2013, the data from micro wave vehicle detection system (MVDs) the big data was obtained in Orlando area by central Florida expressway authority. The paper deals with real time traffic operation and safety monitoring and improvement in expressway.–Random forest and Bayesian inference techniques have been used in real time crash prediction models. –The classification performance is sensitivity of 0.685, specificity of 0.714, accuracy of 0.708 and AUC of 0.755. –The paper proposes a real time congestion and safety monitoring system using MVDs.–The paper can be extended to focus on rear-end crashes.–The performance of the model can be tested in other expressway to assess its potentiality.

9. Deep reinforcement learning for vehicular edge computing: an intelligent offloading system ACM journal ACM Transactions on intelligent systems and technology Not provided –An intelligent offloading system for vehicular edge computing is presented.–Deep reinforcement learning is used for this purpose.–Finite markov chain is used to model communication and computation states. –The offloading request is schedule and network resource allocations are performed using a deep reinforcement learning approach. –A novel deep reinforcement learning based approach is suggested for vehicular edge computing and intelligent offloading system has been constructed.–The problem is formulated as a joint optimization task.–It is reported that the proposed method achieves a matching of 95%and decrease the execution time by 90%.

10. Deep architecture for traffic flow prediction: Deep belief networks with multitask learning IEEE IEEE Transactions on intelligent transportation systems PeMS dataset for traffic flow prediction is used in the study. –The paper develops a promising traffic flow prediction system of transportation modeling and management.–It employs deep belief network (DBN) at the bottom and a multi task regression layer at the top. –The proposed method achieves 5%improved performance compared to existing method.–It also improves the generalization performance. –The hybrid deep learning model extracts required features and performs effective prediction of traffic flow.–This work can be extended by developing temporal deep neural network forecasting model.

11. Leveraging deep learning and SNA approaches for smart city policing in the developing world. Elsevier International Journal of Information Management –The control data record (CDR) which consists of call data record first information report is used.–In addition, citizen profile database is also used.–These datasets pertain to Punjab province. –The paper investigates on smart city policing task. It employs deep learning and social network analysis (SNA) approach. –The proposed approach identifies crime suspects and helps to support smart policing activity. The scheme aims to reduce crime rates in India. –The paper suggests a novel smart city policing scheme using deep learning model.–The proposed method validates its efficiency using real life data.–The potentiality of the model can be tested using more datasets.

12. A novel CNN based security guaranteed image watermarking generation scenario for smart city applications. Elsevier Information Sciences –Standard image datasets. –The paper proposes a CNN based method for image watermarking generation with improved security. –The proposed method performs image processing operations with improved performance. It also detects and extracts watermark. The peak to signal noise is much better than other existing method. –The paper uses a CNN model to classify images.–A classification accuracy of 93.75%is obtained by this method.–The proposed model can be tested with other databases.

13. Nature- inspired algorithm-based secure data dissemination framework for smart city networks. Springer Neural Computing and Applications –Individual paper of this review article employs its own dataset. –The paper deals with from comprehensive review of secured data dissemination for smart community using nature inspired algorithm.–It provides a review of 58 articles on the topic. –The paper suggests three solutions pertaining to smart city network.–Dragon clustering mobility.–Moth flame electric management for smart living and ant colony based intrusion detection system. –The paper provides various solutions for smart city networks. This would assist new researchers to explore and undertake other problems related to smart city.

14. Selection of effective machine learning algorithm and Bot-IoT attacks traffic identification for internet of things in smart city. Elsevier Future Generation Computer Systems –Publicly available online Bot-IoT dataset. (botnet) –The paper develops a cyber attacks detection system for IoT security.–The algorithm used for this purpose is BayesNet, C4.5, NaiveBayes, Random Forest and Random Tree. –The paper develops Machine learning algorithm based attacks identification in IoT network. Out of the five machines learning models random forest performs the best. It is then followed by BayesNet.–However, Naive Bayes is effective for anomaly and intrusion detection. –The experimental results show that the performances of the proposed five algorithms are quite satisfactory. However, the Random forest performs better in terms of accuracy, precision, recall and TP rate.

15. Multiobjective Intelligent energy management for a microgrid. IEEE IEEE Transactions on Industrial Electronics –The data have been recorded in TUAT university and Japan Metrological agency –The paper has proposed energy management for a microgrid using multiobjective approach.–It minimizes the operation cost and environmental impact.–ANN is used for prediction of 24 hours ahead photovoltaic generation and one hour ahead wind power.–Linear programming is used for multiobjective optimization. –The model propose in the paper shows appreciable minimization of operational cost as well as emission level compared to other microgrid energy management systems. –The paper as used multiobjective optimization, ANN based forecasting and fuzzy expert system for energy management of microgrid.–It is reported that satisfactory performance has been achieved.

16. Long-Term wind speed forecasting and general pattern recognition using neural networks. IEEE IEEE Transactions on sustainable energy –The wind speed forecasting data of two metrological station of Malaysia is used. –Different neural networks are used for long term wind speed prediction are employed.–Statistical methods are also used for prediction purpose. –The simulation results obtained using real life data show promising wind speed forecasting with least minimum absolute error. –The authors have suggested neural network based long term wind prediction model. It is demonstrated that the proposed ANN method show improved prediction performance.

17. Deep learning for estimating building energy consumption Elsevier Sustainable Energy, Grids and Networks –Benchmark dataset collected from individual residential customer.–The electric power consumption data is for four years duration with one minute resolution. –The paper proposes using factored conditional restricted Boltzmann machine (FCRBM).–The results are also obtained for SVM, ANN, RNN and Conditional restricted Boltzmann machine (CRBM). –The simulation based results on standard dataset show that the proposed FCRBM method outperforms other four methods used in the study. –The energy consumption in the building is forecasted in the paper using deep learning approach. The proposed FCRBM performs the best compared to other four methods.

18. Learning–based energy-efficient data collection by unmanned vehicles in smart cities IEEE IEEE Transactions on Industrial Informatics –The real datasets are obtained from the mobility traces of taxis in Rome, Italy. It contains GPS coordinates for 320 taxis over 30 days. –In smart cities mobile crowdsourcing (MCS) is having important source of information.–A scheme has been developed to collect this information using unmanned aerial vehicles (UAV).–In this paper, energy efficient MCS collection method is proposed.–Deep reinforcement learning and CNN are used. –The deep reinforcement learning controls the UAV and the CNN extracts the feature and makes decision with the help of deep Q network. –The UAV collects most required data and the mobile unmanned charging station reaches the desired point in shortest time.

19. Human-level control through deep reinforcement learningMacMillan Nature –Atari 2600 platform with a diverse array of task. –The paper has developed a scheme to control human level using deep reinforcement learning. –A deep neural network model has been developed to act as an artificial agent by taking pixel values and the game score as inputs. –The paper provides a high level control performance compared to other existing methods.–The proposed model is shown to be capable enough to perform challenging task.

20. Classifying IoT devices in smart environments using network traffic characteristics IEEE IEEE Transactions on mobile computing –The trace data was obtained using tcpdump tool running on OpenWrt. –The paper develops classification model for IoT devices using traffic characteristics. It employs Naive Bayes and Random forest multidimensional classifiers. –It is shown that the proposed classifiers can identify IoT devices in real time with 99%accuracy has been achieved. –The paper has proposed novel models to classify IoT devices with very high accuracy. The proposed approach makes a comparison between cost performance and speed.

21. Network traffic classifier with convolutional and recurrent neural networks for Internet of Things. IEEE IEEE Access –The source of data is from RedIRIS. It provides RedIRIS which is a Spanish network. It has 500 affiliated institutions. –A novel classifier of network traffic for IoT using convolutional recurrent neural networks is proposed.–The proposed scheme segregates traffic and behavior of heterogeneous devices and services. –It is shown that the proposed approach yields improved detection compared to other models. –The paper proposes several architectures to integrate CNN and RNN. The objective is to classify various IoT traffic. It is also deals with the impact of chosen features and the length of network flow.

22. Student’s affective content analysis in smart classroom environment using deep learning techniques.Springer Multimedia tools and applications –Datasets has been generated by collecting 95–110 videos of Indian students with four different states. –The paper investigates on the analysis of affective content of the students in smart classroom.–The deep learning technique has been employed for this purpose. –The paper classifies four different moods of the students. The experimental results show mood classification accuracy of 90.67%during training and 87.65%during testing phases. –The paper has presented classification of facial expression of students by analyzing affective content.–They have used deep learning approach and have accuracy of 87.65%. The proposed approach can be implemented in real time environment.

23. Virtual learning environment to predict withdrawal by leveraging deep learning.Wiley International Journal of Intelligent Systems –The OULA dataset comprising of demographical information, assessment behavior and log in patterns of 32593 students collected during 2014–2015. –The paper has employed deep LSTM method for predicting the performance of a student.–The investigation predicts early withdrawal of students. –The proposed method shows 97.25%learning accuracy, 85.92%recall and 92.79%precision. –The paper has demonstrated that the prediction of withdrawal under virtual learning environment is better than LR and ANN methods.

24. Multi-objective optimization of building energy performance and indoor thermal comfort: A new method using artificial bee colony (ABC) Elsevier Energy and Buildings –Data related to cold, temperature, warm-dry, warm- humid of four cities of Iran of test rooms have been recorded. –The paper has suggested a multi-objective optimization based approach to obtain a compromise between energy performance and indoor thermal comfort.–The authors have employed artificial bee colony optimization method. –It is reported that the proposed approach minimizes energy consumption by 2.9–11.3%and decrease the dissatisfied percentage by 49.1–56.8%. –The paper has suggested a two- objective optimization problem for evaluating the building performance. The proposed approach provides improved performance compared to single objective based optimization.

25. A hybrid quantum-induced swarm intelligence clustering for the urban trip recommendation in smart city. Elsevier Future Generation Computer Systems –Real time large scale datasets of Yelp and Trip Advisor. –The paper has proposed novel clustering approaches using quantum PSO (QPSO) and quantum induced clustering ensemble (QICE) for recommendation of urban trip in smart city. –The recommendation system developed in this paper has been evaluated using precision, recall, accuracy, F-measure and hit rate. –Yelp dataset-1. QPSO Accuracy- 94.82%, hit rate- 70.64%2. QICEAccuracy- 95.74%, hit rate-76.45%–Trip Advisor dataset-1. QPSO Accuracy- 95.24%, hit rate- 69.59%2. QICEAccuracy- 95.24%, hit rate- 75.68%. The proposed approach outperforms other five methods of clustering.

26. ACO and PSO algorithms for developing a new communication model for VANET applications in smart cities. Springer Wireless personal communication –The data has been obtained from a realistic scenario of traffic simulation. –The paper has proposed a VANET based communication model for smart cities by using ACO and PSO algorithms. –The new communication model has proved its effectiveness by reducing carbon dioxide emission and fuel consumption. –The paper has three contributionsi. New clustering algorithm has been proposed.ii A model to enable fluidity in the warning messages.iii. The use of ACO to find the shortest route.

27. A smart healthcare monitoring system for heart disease prediction based on ensemble deep learning and feature fusion. Elsevier Information Fusion –Cleveland and Hungarian heart disease datasets from UCI. –Development of a smart healthcare system for prediction of heart disease.–It has employed ensemble deep learning approach and fusion of features. –A prediction accuracy of 98.5%which is higher than other methods. –The paper has combined the sensor data and extracted features and then has applied to ensemble deep learning model for prediction of heart disease.–Higher accuracy has been obtained compared to other methods.

28. Deep learning for multigrade brain tumor classification in smart healthcare systems: A prospective survey. IEEE IEEE Transactions on neural networks and learning systems –The datasets are Multigrade brain tumor dataset, brain tumor public dataset, the cancer imaging archieve, BRATS 2015, Harvard and internet brain segmentation repository. –Review of deep learning based classification scheme for multigrade brain tumor. –A systematic review of literature dealing with classification of brain tumor is presented. It deals with pre-processing, feature extraction, classification, results of each method. –The paper has made a survey of deep learning based brain tumor classification. It summarizes the contribution of all the papers and outlines the future research direction

Title of the paper, Names of the Journal and Publisher	Dataset used	Problem statement and methods used	Contribution of the paper	Conclusion
1. Smart energy systems for sustainable smart cities: current developments, trends and future directions.Elsevier Applied Energy	–Individual paper of this review article employs its own dataset.	–The articles related to smart energy systems using CI in smart cities have been reviewed.	–The paper has outlined current developments, the present status and further research scopes.	–The present trend of energy management system using CI has been summarized and presented. It also provides the future direction of energy management in smart cities.
2. Applications of artificial intelligence and machine learning in smart cities. Elsevier Computer Communications	–Individual paper of this review article employs its own dataset.	–The paper has reviewed CI based articles related to energy, transportation, communication, healthcares related to smart city.	–The paper has outlined the present scenario of smart cities in terms of power, communication, healthcare and transport.	–The paper has dealt with shortcomings of smart cities and scope of future development using CI based methods.
3. A survey on an emerging area: Deep learning for smart city data EEE IEEE Transactions on emerging topics in computational intelligence.	–Individual paper of this review article employs its own dataset.	–The papers related to the problems of the smart city using deep learning have been surveyed and the findings are presented.	–It is reported that there are many challenges in smart city which needs solution using deep learning. These have been dealt in the paper in addition to existing solutions	–The security and knowledge fusion issues need to be addressed as future extension of work.
4. A review on deep learning for future smart cities. Wiley Internet Technology Letters	–Individual paper of this review article employs its own dataset.	–A review papers on future smart cities addressing on traffic mobility of citizen and drainage through IoT and using deep learning approach is presented.	–The paper outlines the various deep learning based approaches to solve smart city problems.	–The DL has been applied to the field of transportation, infrastructure, governance, education, healthcare and security.–The challenges are also outlined.
5. Big data analysis in intelligent transportation systems: A surveyIEEE IEEE Transactions on intelligent transportation systems	–Individual paper of this review article employs its own dataset.	–A survey paper on intelligent transportation systems using big data analytics is presented.	–The big data analytics on ITS have provided solution relating to road traffic accident, traffic flow prediction and rail transport management.	–Feasible solutions of transportation systems using intelligent data analytics approaches have been provided. It also outlines open challenges related to big data analytics.
6. Artificial intelligence for smart renewable energy sector in Europe- Smart Energy Infrastructures for next generation smart cities. IEEE IEEE Access	–Individual paper of this review article employs its own dataset.	–The paper provides a summary of review articles on smart energy infrastructure in smart cities of Europe employing CI techniques.	–The paper has analyzed issues related to energy consumption labor productivity and implication of using CI for smart cities.	–The paper outlines the development of a framework in energy sector using CI technique. It also discusses the implications for smart city research and future research scopes.
7. LSTM network: a deep learning approach for short-term traffic forecast. IET IET Intelligent transport systems	Traffic data collected from Beijing traffic management bureau	Forecasting of short term traffic.–A deep LSTM network is used.	–The proposed LSTM model achieves better short term forecasting of traffic compared to ARIMA, SVM, RBF, SAE and RNN method.	–The paper has developed LSTM based traffic volume prediction model. The work can be extended to build multiple output traffic forecast system.
8. Big data applications in real-time traffic operation and safety monitoring and improvement on urban expressways. Elsevier Transportation Research Part C	In 2013, the data from micro wave vehicle detection system (MVDs) the big data was obtained in Orlando area by central Florida expressway authority.	The paper deals with real time traffic operation and safety monitoring and improvement in expressway.–Random forest and Bayesian inference techniques have been used in real time crash prediction models.	–The classification performance is sensitivity of 0.685, specificity of 0.714, accuracy of 0.708 and AUC of 0.755.	–The paper proposes a real time congestion and safety monitoring system using MVDs.–The paper can be extended to focus on rear-end crashes.–The performance of the model can be tested in other expressway to assess its potentiality.
9. Deep reinforcement learning for vehicular edge computing: an intelligent offloading system ACM journal ACM Transactions on intelligent systems and technology	Not provided	–An intelligent offloading system for vehicular edge computing is presented.–Deep reinforcement learning is used for this purpose.–Finite markov chain is used to model communication and computation states.	–The offloading request is schedule and network resource allocations are performed using a deep reinforcement learning approach.	–A novel deep reinforcement learning based approach is suggested for vehicular edge computing and intelligent offloading system has been constructed.–The problem is formulated as a joint optimization task.–It is reported that the proposed method achieves a matching of 95%and decrease the execution time by 90%.
10. Deep architecture for traffic flow prediction: Deep belief networks with multitask learning IEEE IEEE Transactions on intelligent transportation systems	PeMS dataset for traffic flow prediction is used in the study.	–The paper develops a promising traffic flow prediction system of transportation modeling and management.–It employs deep belief network (DBN) at the bottom and a multi task regression layer at the top.	–The proposed method achieves 5%improved performance compared to existing method.–It also improves the generalization performance.	–The hybrid deep learning model extracts required features and performs effective prediction of traffic flow.–This work can be extended by developing temporal deep neural network forecasting model.
11. Leveraging deep learning and SNA approaches for smart city policing in the developing world. Elsevier International Journal of Information Management	–The control data record (CDR) which consists of call data record first information report is used.–In addition, citizen profile database is also used.–These datasets pertain to Punjab province.	–The paper investigates on smart city policing task. It employs deep learning and social network analysis (SNA) approach.	–The proposed approach identifies crime suspects and helps to support smart policing activity. The scheme aims to reduce crime rates in India.	–The paper suggests a novel smart city policing scheme using deep learning model.–The proposed method validates its efficiency using real life data.–The potentiality of the model can be tested using more datasets.
12. A novel CNN based security guaranteed image watermarking generation scenario for smart city applications. Elsevier Information Sciences	–Standard image datasets.	–The paper proposes a CNN based method for image watermarking generation with improved security.	–The proposed method performs image processing operations with improved performance. It also detects and extracts watermark. The peak to signal noise is much better than other existing method.	–The paper uses a CNN model to classify images.–A classification accuracy of 93.75%is obtained by this method.–The proposed model can be tested with other databases.
13. Nature- inspired algorithm-based secure data dissemination framework for smart city networks. Springer Neural Computing and Applications	–Individual paper of this review article employs its own dataset.	–The paper deals with from comprehensive review of secured data dissemination for smart community using nature inspired algorithm.–It provides a review of 58 articles on the topic.	–The paper suggests three solutions pertaining to smart city network.–Dragon clustering mobility.–Moth flame electric management for smart living and ant colony based intrusion detection system.	–The paper provides various solutions for smart city networks. This would assist new researchers to explore and undertake other problems related to smart city.
14. Selection of effective machine learning algorithm and Bot-IoT attacks traffic identification for internet of things in smart city. Elsevier Future Generation Computer Systems	–Publicly available online Bot-IoT dataset. (botnet)	–The paper develops a cyber attacks detection system for IoT security.–The algorithm used for this purpose is BayesNet, C4.5, NaiveBayes, Random Forest and Random Tree.	–The paper develops Machine learning algorithm based attacks identification in IoT network. Out of the five machines learning models random forest performs the best. It is then followed by BayesNet.–However, Naive Bayes is effective for anomaly and intrusion detection.	–The experimental results show that the performances of the proposed five algorithms are quite satisfactory. However, the Random forest performs better in terms of accuracy, precision, recall and TP rate.
15. Multiobjective Intelligent energy management for a microgrid. IEEE IEEE Transactions on Industrial Electronics	–The data have been recorded in TUAT university and Japan Metrological agency	–The paper has proposed energy management for a microgrid using multiobjective approach.–It minimizes the operation cost and environmental impact.–ANN is used for prediction of 24 hours ahead photovoltaic generation and one hour ahead wind power.–Linear programming is used for multiobjective optimization.	–The model propose in the paper shows appreciable minimization of operational cost as well as emission level compared to other microgrid energy management systems.	–The paper as used multiobjective optimization, ANN based forecasting and fuzzy expert system for energy management of microgrid.–It is reported that satisfactory performance has been achieved.
16. Long-Term wind speed forecasting and general pattern recognition using neural networks. IEEE IEEE Transactions on sustainable energy	–The wind speed forecasting data of two metrological station of Malaysia is used.	–Different neural networks are used for long term wind speed prediction are employed.–Statistical methods are also used for prediction purpose.	–The simulation results obtained using real life data show promising wind speed forecasting with least minimum absolute error.	–The authors have suggested neural network based long term wind prediction model. It is demonstrated that the proposed ANN method show improved prediction performance.
17. Deep learning for estimating building energy consumption Elsevier Sustainable Energy, Grids and Networks	–Benchmark dataset collected from individual residential customer.–The electric power consumption data is for four years duration with one minute resolution.	–The paper proposes using factored conditional restricted Boltzmann machine (FCRBM).–The results are also obtained for SVM, ANN, RNN and Conditional restricted Boltzmann machine (CRBM).	–The simulation based results on standard dataset show that the proposed FCRBM method outperforms other four methods used in the study.	–The energy consumption in the building is forecasted in the paper using deep learning approach. The proposed FCRBM performs the best compared to other four methods.
18. Learning–based energy-efficient data collection by unmanned vehicles in smart cities IEEE IEEE Transactions on Industrial Informatics	–The real datasets are obtained from the mobility traces of taxis in Rome, Italy. It contains GPS coordinates for 320 taxis over 30 days.	–In smart cities mobile crowdsourcing (MCS) is having important source of information.–A scheme has been developed to collect this information using unmanned aerial vehicles (UAV).–In this paper, energy efficient MCS collection method is proposed.–Deep reinforcement learning and CNN are used.	–The deep reinforcement learning controls the UAV and the CNN extracts the feature and makes decision with the help of deep Q network.	–The UAV collects most required data and the mobile unmanned charging station reaches the desired point in shortest time.
19. Human-level control through deep reinforcement learningMacMillan Nature	–Atari 2600 platform with a diverse array of task.	–The paper has developed a scheme to control human level using deep reinforcement learning.	–A deep neural network model has been developed to act as an artificial agent by taking pixel values and the game score as inputs.	–The paper provides a high level control performance compared to other existing methods.–The proposed model is shown to be capable enough to perform challenging task.
20. Classifying IoT devices in smart environments using network traffic characteristics IEEE IEEE Transactions on mobile computing	–The trace data was obtained using tcpdump tool running on OpenWrt.	–The paper develops classification model for IoT devices using traffic characteristics. It employs Naive Bayes and Random forest multidimensional classifiers.	–It is shown that the proposed classifiers can identify IoT devices in real time with 99%accuracy has been achieved.	–The paper has proposed novel models to classify IoT devices with very high accuracy. The proposed approach makes a comparison between cost performance and speed.
21. Network traffic classifier with convolutional and recurrent neural networks for Internet of Things. IEEE IEEE Access	–The source of data is from RedIRIS. It provides RedIRIS which is a Spanish network. It has 500 affiliated institutions.	–A novel classifier of network traffic for IoT using convolutional recurrent neural networks is proposed.–The proposed scheme segregates traffic and behavior of heterogeneous devices and services.	–It is shown that the proposed approach yields improved detection compared to other models.	–The paper proposes several architectures to integrate CNN and RNN. The objective is to classify various IoT traffic. It is also deals with the impact of chosen features and the length of network flow.
22. Student’s affective content analysis in smart classroom environment using deep learning techniques.Springer Multimedia tools and applications	–Datasets has been generated by collecting 95–110 videos of Indian students with four different states.	–The paper investigates on the analysis of affective content of the students in smart classroom.–The deep learning technique has been employed for this purpose.	–The paper classifies four different moods of the students. The experimental results show mood classification accuracy of 90.67%during training and 87.65%during testing phases.	–The paper has presented classification of facial expression of students by analyzing affective content.–They have used deep learning approach and have accuracy of 87.65%. The proposed approach can be implemented in real time environment.
23. Virtual learning environment to predict withdrawal by leveraging deep learning.Wiley International Journal of Intelligent Systems	–The OULA dataset comprising of demographical information, assessment behavior and log in patterns of 32593 students collected during 2014–2015.	–The paper has employed deep LSTM method for predicting the performance of a student.–The investigation predicts early withdrawal of students.	–The proposed method shows 97.25%learning accuracy, 85.92%recall and 92.79%precision.	–The paper has demonstrated that the prediction of withdrawal under virtual learning environment is better than LR and ANN methods.
24. Multi-objective optimization of building energy performance and indoor thermal comfort: A new method using artificial bee colony (ABC) Elsevier Energy and Buildings	–Data related to cold, temperature, warm-dry, warm- humid of four cities of Iran of test rooms have been recorded.	–The paper has suggested a multi-objective optimization based approach to obtain a compromise between energy performance and indoor thermal comfort.–The authors have employed artificial bee colony optimization method.	–It is reported that the proposed approach minimizes energy consumption by 2.9–11.3%and decrease the dissatisfied percentage by 49.1–56.8%.	–The paper has suggested a two- objective optimization problem for evaluating the building performance. The proposed approach provides improved performance compared to single objective based optimization.
25. A hybrid quantum-induced swarm intelligence clustering for the urban trip recommendation in smart city. Elsevier Future Generation Computer Systems	–Real time large scale datasets of Yelp and Trip Advisor.	–The paper has proposed novel clustering approaches using quantum PSO (QPSO) and quantum induced clustering ensemble (QICE) for recommendation of urban trip in smart city.	–The recommendation system developed in this paper has been evaluated using precision, recall, accuracy, F-measure and hit rate.	–Yelp dataset-1. QPSO Accuracy- 94.82%, hit rate- 70.64%2. QICEAccuracy- 95.74%, hit rate-76.45%–Trip Advisor dataset-1. QPSO Accuracy- 95.24%, hit rate- 69.59%2. QICEAccuracy- 95.24%, hit rate- 75.68%. The proposed approach outperforms other five methods of clustering.
26. ACO and PSO algorithms for developing a new communication model for VANET applications in smart cities. Springer Wireless personal communication	–The data has been obtained from a realistic scenario of traffic simulation.	–The paper has proposed a VANET based communication model for smart cities by using ACO and PSO algorithms.	–The new communication model has proved its effectiveness by reducing carbon dioxide emission and fuel consumption.	–The paper has three contributionsi. New clustering algorithm has been proposed.ii A model to enable fluidity in the warning messages.iii. The use of ACO to find the shortest route.
27. A smart healthcare monitoring system for heart disease prediction based on ensemble deep learning and feature fusion. Elsevier Information Fusion	–Cleveland and Hungarian heart disease datasets from UCI.	–Development of a smart healthcare system for prediction of heart disease.–It has employed ensemble deep learning approach and fusion of features.	–A prediction accuracy of 98.5%which is higher than other methods.	–The paper has combined the sensor data and extracted features and then has applied to ensemble deep learning model for prediction of heart disease.–Higher accuracy has been obtained compared to other methods.
28. Deep learning for multigrade brain tumor classification in smart healthcare systems: A prospective survey. IEEE IEEE Transactions on neural networks and learning systems	–The datasets are Multigrade brain tumor dataset, brain tumor public dataset, the cancer imaging archieve, BRATS 2015, Harvard and internet brain segmentation repository.	–Review of deep learning based classification scheme for multigrade brain tumor.	–A systematic review of literature dealing with classification of brain tumor is presented. It deals with pre-processing, feature extraction, classification, results of each method.	–The paper has made a survey of deep learning based brain tumor classification. It summarizes the contribution of all the papers and outlines the future research direction

It is observed from the review of articles that the researchers are working in nine different aspects of smart city community. These are listed in the beginning of Section 2. Accordingly the categorization of the articles have been made and presented in Table 2. It is observed from this table that the numbers of articles in the nine categories are 6, 4, 4, 3, 2, 2, 2, 3 and 2 respectively. The table reveals that maximum numbers of articles are being published in the sub area of literature review and intelligent transportation system.

Table 2

Categorization of articles

S. No	Category	Reference number of articles	S. No	Category	Reference number of articles
1	Review articles on smart city (RA)	[1 –6]	6	SE	[20–21]
2	ITS	[7 –10]	7	SEd	[22]–
3	Security	[11 –14]	8	SI	[24 –26]
4	Smart energy	[15 –17]	9	SH	[27–28]
5	UAV	[18–19]

4 Sources of data and CI techniques used

In this Section, the standard sources of data pertaining to smart city community used by authors of different articles have been collected and presented in Table 3. Many of the data sets have been obtained from real life experiments and others have been taken from standard websites.

Table 3
Sources of data

References Sources of data References Sources of data

[7] Traffic data collected from Beijing traffic management bureau [21] The source of data is from RedIRIS. It provides RedIRIS which is a Spanish network. It has 500 affiliated institutions.

[8] In 2013, the data from micro wave vehicle detection system (MVDs) the big data was obtained in Orlando area by central Florida expressway authority. [22] Datasets has been generated by collecting 95–110 videos of Indian students with four different states.

[10] PeMS dataset for traffic flow prediction is used in the study. [23] The OULA dataset comprising of demographical information, assessment behavior and log in patterns of 32593 students collected during 2014–2015.

[11] –The control data record (CDR) which consists of call data record first information report is used.–In addition, citizen profile database is also used.–These datasets pertain to Punjab province. [24] Data related to cold, temperature, warm-dry, warm-humid of four cities of Iran of test rooms have been recorded.

[12] Standard image datasets. [25] Real time large scale datasets of Yelp and Trip Advisor.

[14] Publicly available onlineBot-IoT dataset. (botnet) [26] The data has been obtained from a realistic scenario of traffic simulation.

[15] The data have been recorded in TUAT university and Japan Metrological agency [27] Cleveland and Hungarian heart disease datasets from UCI.

[16] The wind speed forecasting data of two metrological station of Malaysia are used. [28] The datasets are Multigrade brain tumor dataset, brain tumor public dataset, the cancer imaging archieve, BRATS 2015, Harvard and internet brain segmentation repository.

[17] Benchmark dataset collected from individual residential customer.–The electric power consumption data is for four years duration with one minute resolution.

[18] The real datasets are obtained from the mobility traces of taxis in Rome, Italy. It contains GPS coordinates for 320 taxis over 30 days.

[19] Atari 2600 platform with a diverse array of task.

[20] The trace data was obtained using tcpdump tool running on OpenWrt.

References	Sources of data	References	Sources of data
[7]	Traffic data collected from Beijing traffic management bureau	[21]	The source of data is from RedIRIS. It provides RedIRIS which is a Spanish network. It has 500 affiliated institutions.
[8]	In 2013, the data from micro wave vehicle detection system (MVDs) the big data was obtained in Orlando area by central Florida expressway authority.	[22]	Datasets has been generated by collecting 95–110 videos of Indian students with four different states.
[10]	PeMS dataset for traffic flow prediction is used in the study.	[23]	The OULA dataset comprising of demographical information, assessment behavior and log in patterns of 32593 students collected during 2014–2015.
[11]	–The control data record (CDR) which consists of call data record first information report is used.–In addition, citizen profile database is also used.–These datasets pertain to Punjab province.	[24]	Data related to cold, temperature, warm-dry, warm-humid of four cities of Iran of test rooms have been recorded.
[12]	Standard image datasets.	[25]	Real time large scale datasets of Yelp and Trip Advisor.
[14]	Publicly available onlineBot-IoT dataset. (botnet)	[26]	The data has been obtained from a realistic scenario of traffic simulation.
[15]	The data have been recorded in TUAT university and Japan Metrological agency	[27]	Cleveland and Hungarian heart disease datasets from UCI.
[16]	The wind speed forecasting data of two metrological station of Malaysia are used.	[28]	The datasets are Multigrade brain tumor dataset, brain tumor public dataset, the cancer imaging archieve, BRATS 2015, Harvard and internet brain segmentation repository.
[17]	Benchmark dataset collected from individual residential customer.–The electric power consumption data is for four years duration with one minute resolution.
[18]	The real datasets are obtained from the mobility traces of taxis in Rome, Italy. It contains GPS coordinates for 320 taxis over 30 days.
[19]	Atari 2600 platform with a diverse array of task.
[20]	The trace data was obtained using tcpdump tool running on OpenWrt.

Most of the articles have focused on obtaining appropriate decision relating to smart city community by employing machine learning and CI techniques. These have been identified from each article and presented in Table 4.

Table 4

CI Techniques used for different tasks related to smart city community

Category	References	Tasks	CI method employed
ITS	[7]	Short –term traffic
forecast	LSTM
	[8]	Road traffic accident	Random Forest
	[9]	Offloading System	Deep Reinforcement Learning
	[10]	Road traffic flow prediction	Deep Belief Network
Security	[11]	Smart city policing (classification between criminals and non criminals)	Deep learning
	[12]	Security guaranteed image watermarking	CNN
	[13]	1. Clustering mobility in IoV	Dragon fly algorithm
		2. Electric management for smart living	Moth flame algorithm
		3. Intrusion detection system	Ant colony algorithm
	[14]	Identification of attacks in IoT networks.	BayesNet, C4.5, NaiveBayes, Random Forest, Random tree
SE	[15]	Energy management of microgrid	1. ANN Ensemble is used to predict 24 hour ahead photovoltaic generation.2. Fuzzy expert system is used for battery scheduling.3. Multiobjective energy management using linear programming based multiobjective optimization.
	[16]	Long-term wind speed forecasting and pattern recognition	ANN is used for wind speed forecasting.
	[17]	Estimation of building energy consumption	Prediction of building energy consumption by ANN, RNN, SVM, conditional restricted Boltzmann machine (CRBM) and Factored CRBM (FCRBM)
UAV	[18]	Energy efficient data collected by UAV	CNN for classification
	[19]	Human level control	Deep reinforcement learning
SE	[20]	Classification of IoT devices in smart environment using network traffic characteristics	1. Naive Bayes classifier2. Random forest
	[21]	Network traffic classifier for IoT	CNN and RNN
SEd	[22]	Content analysis in smart classroom for mood classification	Deep learning
	[23]	Prediction of withdrawal in virtual learning	Deep learning
SI	[24]	Optimization of building energy performance and in door thermal comfort	Artificial bee colony optimization
	[25]	Clustering for urban trip recommendation	Hybrid quantum induced swarm intelligence
	[26]	Communication model for vehicular adhoc network (VANET)	ACO and PSO algorithm
SH	[27]	Smart healthcare monitoring system (prediction of heart disease)	Feature fusion and deep learning approaches
	[28]	Classification of brain tumor in smart healthcare system	Deep learning

5 Brief outline of different CI methods used

This Section deals with the various CI methods used in problems related to smart city community. It provides brief outline of each of the CI methods listed in Table 5.

Table 5
List of various CI methods used in smart city community articles

References Techniques References Techniques

[1 –6] Review [14] Random tree

[7] LSTM [15, 16], [17] ANN Ensemble

[8 , 20] RF [15] FES

[9, 19] Deep reinforcement learning [15] Multi objective optimization

[10] Deep belief network [17, 21] RNN

[10] Multi regression [17] SVM

[11, 22], [23 , 28] Deep learning [17] CRBM

[12 , 21] CNN [17] FCRBM

[13] Dragon fly [24] ABC

[13] Moth flame [25] Hybrid quantum induced swarm intelligence

[13, 26] ACO [26] PSO

[14] BayesNet [27] Feature fusion

[14] C4.5

[14, 20] Naive Bayes

References	Techniques	References	Techniques
[1 –6]	Review	[14]	Random tree
[7]	LSTM	[15, 16], [17]	ANN Ensemble
[8 , 20]	RF	[15]	FES
[9, 19]	Deep reinforcement learning	[15]	Multi objective optimization
[10]	Deep belief network	[17, 21]	RNN
[10]	Multi regression	[17]	SVM
[11, 22], [23 , 28]	Deep learning	[17]	CRBM
[12 , 21]	CNN	[17]	FCRBM
[13]	Dragon fly	[24]	ABC
[13]	Moth flame	[25]	Hybrid quantum induced swarm intelligence
[13, 26]	ACO	[26]	PSO
[14]	BayesNet	[27]	Feature fusion
[14]	C4.5
[14, 20]	Naive Bayes

5.1 Long short-term memory (LSTM)

The Long short-term memory (LSTM), proposed by Gers et al. in 2000, is a feed forward-feedback artificial recurrent neural network. It processes single data point as well as sequence of data. The LSTM is superior to recurrent neural network (RNN) in terms of control ability. There are five steps in LSTM model. The first step defines the network as a sequence of layers. The LSTM layer comprises of memory cells. The second step, transforms the simple sequence of layers into a format which can be executed by GPU. In the third step, the weights are adapted using back propagation learning. In fourth step, after completion of successful training the performance of the network is evaluated. The LSTM model retains the information for a long period of time and hence in fifth step, the trained model is used for predicting and classifying tasks as per the requirement.

5.2 Random forest (RF)

The random forest, proposed by Tin Kam Ho, and developed by Leo and Adele in 2019 is an ensemble learning method for classification and prediction operation. It constructs a multiple of decision trees during training time. Finally, the class is estimated on the mean prediction of the individual trees. The overfitting issue associated with decision trees is overcome by random forest. The random forest outperforms the decision trees. The steps involved in random forest algorithm is-

Select random samples from a given dataset

Construct a separate decision tree for every randomly selected set.

Obtain prediction result from each decision tree.

Perform voting for every predicted result

Choose the final prediction result by selecting the most voted prediction.

5.3 Deep reinforcement learning (DRL)

It combines the principle of both deep and reinforcement learnings to exploit the benefits of the two. In reinforcement learning an agent learns to perform an action through the method of trial and error. Based on the reward received by the agent it optimizes its behavior until the goal is achieved. Through DRL the desired output is achieved by passing the raw data through number of layers of reinforcement learning.

5.4 Deep belief network (DBN)

The DBN is a generative graphical model which learns to reconstruct its inputs. It comprises of multiple layers of hidden units such that there is no connections between units within each layer but there is connection between the layers. The layers of DBN act as feature detectors. After the completion of the training stage, the DBN can further learn with supervision to function as a classifier. The DBN can be viewed as a deep learning model as it is trained greedily one layer at a time. The steps involved in the training of the DBN are:

Initialize the training vector (Visible units)

Update the hidden units in parallel.

Update the visible units in parallel.

Reupdate the hidden units.

Update the weights.

The DBN finds applications in classification and forecasting tasks in many engineering fields.

5.5 Multiple Regressions (MR)

It establishes the straight line relationships among two or more variables. The MR estimates the regression coefficients by least square approach. The steps employed in MR based approach are:

Model building

Model Adequacy

Model Assumptions

Potential Modeling and solutions

Model validation

5.6 Deep learning (DL)

The DL algorithms are a subset of machine learning algorithms which uses multiple layers to sequentially extract higher level features from the raw input data. The deep refers to the number of layers through which the data is transformed. It helps to pick up the right features which improve the prediction and classification performance. The DL can be applied for speech and audio recognition, bio-informatics, drug design, computer vision, health care etc. The major seven steps involved in DL are:

Collect the raw data

Pre-process the data

Choose a deep learning model

Train the model

Evaluate the model

Tune the hyperparameters

Perform prediction or classification task

5.7 Convolutional neural network (CNN)

The CNN is a class of DL. There also known as space invariant artificial neural network (SIANN). The CNN comprises of multiple perceptrons in which networks are fully connected. The hidden layers perform series of convolution. The ReLu function is commonly used as activation function. The pooling operations are carried out to reduce the dimensions of the data without much sacrificing the quality of the data. After suitable features are extracted the two dimensional convolved data as converted to one dimension before the prediction or classification operation is carried out in last few neural networks.

5.8 Dragonfly algorithm (DA)

Based on the dynamic and static swarming behavior of artificial dragon flies the DA algorithm has been developed in the literature. The various step involved in this algorithm are:

Separation

It indicates the static collision avoidance in which the individuals follow to avoid collision with each other.

Alignment

It represents the velocity matching of individual with other neighborhood of the same group.

Cohesion

It represents the tendency of the individuals towards the centre of the group.

Attraction

It denotes the attraction towards the food source

Distraction

It computes the diversion from the enemy.

It performs better than several meta-heuristic algorithms.

5.9 Moth-flame optimization (MFO)

The MFO, Proposed by Mirjalili, is one of the promising meta-heuristic algorithms which finds application in many fields. The steps involved in MFO are:

Define the parameters of the algorithm

Generate the initial moths randomly

Calculate the fitness functions and find the best positions by flames.

Update the flame number

Calculate the space between the ith moth and jth flame.

Update position of each moth.

Move to step 3, if terminating criteria are not satisfied.

End

This MFO finds applications in various optimization problems, data mining, clustering and medical engineering.

5.10 Ant colony optimization (ACO)

The ACO, proposed by Dorigo in 1992 is an important optimization algorithm which has been developed by following the behavior of the ants. The ACO is applied to vehicle routing, internet routing, travelling salesman problem (TSP) and other commercial applications. The principle behind ACO is computation of pheromone intensity of ants through the path on which each ant travels. An artificial ant travels towards a node based on a path on which the probability is highest. The steps of ACO are:

Initialization

Initialize pheromone intensity in each path

Initialize the change in pheromone intensity in each path

Compute inverse of the given distance matrix.

Compute the probability of each path using Steps 1 and 2.

Place the ant at a node which leads through a path of maximum probability.

Update the change in pheromone intensity as well as pheromone intensity of each path.

Repeat Steps 3 to 5 for the same ant until it reaches to the initial node.

Repeat Steps 3 to 6 for each ant

Find the best route (which offers the least distance) by comparing the minimum of the distances travelled by all artificial ants.

The least distance provides the solution.

5.11 Bayesian network (BN)

It is also called belief network or Bayes Network. It represents a probabilistic graphical model based on directed acyclic graph. This network establishes probabilistic relationship between cause and effect. The steps involved in this network are:

Identification of important model parameters which are required for estimating the risk of violent re-offence.

Construction of casual model structure using the variables identified in Step 1

Linking of the required data to variables of models.

Performance of parameterization of model and handling of missing data using expectation maximization algorithm.

Review of the behavior of the developed model and suggestion for further revision if required.

The BN is a popular method for modeling uncertain and complexed systems. It provides robust framework for the analysis of the problem.

5.12 C_4.5 algorithm

It is an extension of ID₃ algorithm and known as statistical classifier. It generates decision trees which are used for classification. It is a widely used as an efficient classifier. The steps involved in this algorithm are:

Construct a node called N

Return N as a leaf node, labeled with C (If all the training dataset belongs to same class C)

Return N as a leaf node labeled with majority class if attribute list is empty.

The C_4.5 has few improvements over ID₃. These are:

It is capable of handling both discrete and continuous variables.

It can handle training data with missing attribute values.

It can handle attributes with different costs.

5.13 Naive bayes (NB)

The NB is a probabilistic classifier which employs Bayes’ theorem. This classifier is highly scalable, requires a number of linear parameters. It is based on maximum-likelihood training and does not employ iterative method of learning. The Steps in NB classifier are:

Compute the frequency table using the given datasets

Create likelihood table using Step 1.

Calculate the posterior probability for each class using Naive Bayes equation.

Estimate the class based on highest posterior probability.

It finds the class faster. It performs well in the categorical input. One limitation of this classifier is the assumption of the independent predictor.

5.14 Random tree

A random tree represents a spanning tree of a graph in which each tree has equal chance of being selected. Two different distributions are usually used. According to a random permutation binary trees are formed by inserting one node at a time. In the other case, binary trees are chosen from uniform discrete distribution.

5.15 ANN Ensemble (ANNE)

The ANN Ensemble is a two-stage model which includes model training and model combination. During training phase each ANN model is trained with same input set. Each model is then used to make a prediction or classification. The final result is obtained by taking the average of the results. In some cases, the output of each of the ANN model is weighted and then summed to produce the final output. The magnitude of each of the weight is found by employing biologically inspired technique based optimization. The key steps of ANNE model are:

Vary the choice of data to train the each ANN model

Vary the choice of the models used in the ensemble system.

Vary the choice of the way that the outputs from ensemble models are combined.

5.16 Fuzzy expert system (FES)

The FES uses fuzzy logic. It comprises of fuzzification, inference, composition and de-fuzzification. In the inference process using rules and membership functions, the output variables are computed using specific values of input variables. In the fuzzification stage, using suitable membership function, the input values are mapped to obtain the degree of truth. In the inference stage the truth value is computed and applied to each rule. In the composition stage all the fuzzy subsets are assigned to each output variable are combined together. Then a single fuzzy subset for each output variable is obtained. In the de-fuzzification stage, the membership values are converted back to the crisp values. There are many methods of defuzzification such as centroid method and average maximum method. The defuzzified output provides the desired result. The FESs is used in several fields such as pattern recognition, non-linear control and financial systems.

5.17 Multi-objective optimization (MOO)

It refers to optimization of multi objective problems with many variables and many constraints. Mostly the MOO problems are solved using biologically inspired techniques. The MOO finds applications in almost all fields for the purpose of prediction, classification, detection tasks. Some common types of biologically inspired MOOs are Non-dominated sorting algorithm (NSGA) version 2, Multi-objective particle swarm optimization (MOPSO), Multi-objective cat swarm optimization (MOCSO) etc. These algorithms are population based and work iteratively to arrive at an optimized solution. Even though the process is time taking, the solutions are better and can be conveniently used for offline applications.

5.18 Recurrent neural network (RNN)

The RNN is a generalization of artificial neural network having internal memory. In RNN the same function is performed for every input of the data. The output of the current input depends on the past one computation. After the output is produced it is sent back to the recurrent network to arrive at a decision. It uses the current input and the output which has been obtained from the previous input. The RNN models sequence of data so that each sample is assumed to be dependent on previous one. The major steps in RNN are:

Apply an input at a particular time.

Compute its current state using current input and previous state.

Assume the current state as the previous state for the next time state.

Repeat steps 1 to 3 until all time steps are completed and then the output is computed using the final current state.

Generate the error term by comparing the actual and target outputs.

Back propagate the error to update the connecting weights.

The RNN is associated with gradient vanishing and exploding problems.

5.19 Support vector machine (SVM)

The SVM algorithm was invented by Vapnik and Chervonenkis in 1963 for non linear classification purpose. The SVM model represents the input datasets as points in the space in such a way that the data are divided into separate categories by a maximum possible clear gap. The steps of SVM are:

Prepare and format the dataset.

Normalize the datasets

Select activation function

Optimize the parameters using search algorithm

Train the SVM model

Test SVM model

Evaluate the performance of classification

The SVM is more effective if dimensional spaces are high and particularly number of dimension is more than the number of samples.

5.20 Conditional restricted boltzmann machine (CRBM)

The CRBM, introduced by Taylor and Hinton in 2009, is used for multi dimensional system modeling which learns using the probability distribution over a set of data. It consists of history, hidden and present layers. These layers are connected through a three way weight tensor among them. When an input dataset is applied the CRBM learns by fitting the weight tensor in such a way that the energy function is progressively minimized. The training of the algorithm is computationally expensive.

5.21 Factored conditional restricted boltzmann machine (FCRBM)

The RBMs are unidirectional graphical models which use hidden variables to develop higher order non linear prediction models. The RBM represents probability distributions over random variables under an energy based model. After completion of training the model it provides a closed form representation for the distribution following the observation. The CRBM model is suitable for multi-label classification given a training dataset with missing labels. The true label vectors are simultaneously learnt by maximizing the regularized conditional marginal likelihood of the label. In addition, it incorporates label co-occurrence information received from auxiliary resources as previous knowledge. The FCRBM is a compact and efficient multi class model. It is a deep layered structure and employs ReLu function and multi variate auto regressive algorithm for training. The performance of this model is consistent and robust in nature.

5.22 Artificial bee colony (ABC) algorithm

It is a bio-inspired optimization algorithm which exploits foraging behavior of honey bee. It was proposed by Dervis Karaboga in 2005. The colony comprises of employed bees onlookers and scouts. It is assumed that the number of food sources surrounding the hive is equal to the number of employed bees. The employed bees after returning from food source dance near the hive. The employed bee becomes a scout after its food source is abandoned. It is then searches for new food source. The onlookers choose the food sources based on the dances of the employed bees. The steps of the algorithms are:

Produce the initial food sources for all employed bees.

Each employed bee finds a closest food source based on her memory.

It evaluates its nectar amount and dances in the hive.

Each onlooker chooses a source depending on the dances of the employed bee. It goes to the source and evaluates the nectar amount.

The scouts discover the abandoned food sources which are replaced by new food sources.

The best food source is identified.

Steps 2 to 6 are repeated until the condition in step 6 is satisfied.

The ABC is a promising optimization algorithm which finds potential applications in many fields.

5.23 Quantum induced swarm intelligence (QPSO)

It is a new swarm intelligence algorithm, introduced in 2004 with improved ability over PSO. It involves fewer adjustable parameters and produces effective solution better than the PSO. It assumes that the particle in the space has quantum behavior. The steps of the algorithm are:-

Initialize the population

Initialize the optimal particle history and global optimal history.

Evaluate the fitness function of particles.

Update history optimal of particles.

Update the global optimum of particle swarm history.

Update all particles in the population.

Check the terminating condition.

Go to step 2 if step 7 is not satisfied. Otherwise outputs the optimal solution.

This algorithm is a powerful one and provides better and faster solution in many applications.

5.24 Particle Swarm Optimization (PSO)

It is a population based optimization algorithm originally contributed by Kennedy, Eberhart and Shi. Each member of the population is called a particle and a set of particles constitutes a swarm. Each particle has a position as well as a velocity. The initial position for each particle represents an initial solution. Each particle has personal best (p-best) value which changes from time to time. At a given time, the best of the p-best is known as global best (g-best) which represents the solution. After each time instant the velocity and the position of each particle are updated using the p-best of each particle as well as the g-best. The steps of PSO algorithm are:

Initialize the position and velocity of each particle of the swarm.

Compute the p-best of each particle and the g-best of the swarm.

Update the velocity and position of each particle using the p-best and g-best values.

Update the p-best and g-best.

Repeat Steps 3 and 4 until the stopping criterion is made (The value of g-best remains constant).

The PSO is a simple but effective optimization algorithm which finds applications in all fields of the engineering.

5.25 Feature fusion

The feature fusion learns the features of the given dataset which are compact representation of integrated features. It involves lower computational complexity and provides better detection performance. It integrates multiple different feature information to obtain more prominent set of features.

6 Analysis of results presented in various articles

The reviews of the key results in the collected articles have been made and are presented in Table 6. This table also presents the best possible performance achieved by the CI method employed

Table 6
Analysis of results

References Performance metrics Results Referencees Performance metrics Results

[7] Mean Relative Error (MRE) 15 minPoint A-6.41 Point B-6.05 Point C-6.21 30 minPoint A-9.45 Point B-10.01 Point C-8.67 60 minPoint A-16.25 Point B-17.20 Point C-17.63 [22] Accuracy (during training) 90.67%

Accuracy (during testing) 87.65%

[8] Sensitivity 0.685 [23] Accuracy (learni ng accuracy) 97.25%

Specificity 0.714

Accuracy 0.708 Recall 85.92%

AUC 0.755 Precision 92.79%

[12] Accuracy 93.75% [25] Average max accuracy QPSO-94.93%

QICE-95.91%

[20] Accuracy 99% [28] Accuracy 98.5%

References	Performance metrics	Results	Referencees	Performance metrics	Results
[7]	Mean Relative Error (MRE)	15 minPoint A-6.41 Point B-6.05 Point C-6.21 30 minPoint A-9.45 Point B-10.01 Point C-8.67 60 minPoint A-16.25 Point B-17.20 Point C-17.63	[22]	Accuracy (during training)	90.67%
				Accuracy (during testing)	87.65%
[8]	Sensitivity	0.685	[23]	Accuracy (learni ng accuracy)	97.25%
	Specificity	0.714
	Accuracy	0.708		Recall	85.92%
	AUC	0.755		Precision	92.79%
[12]	Accuracy	93.75%	[25]	Average max accuracy	QPSO-94.93%
					QICE-95.91%
[20]	Accuracy	99%	[28]	Accuracy	98.5%

The observation of Table 6 shows the name of the CI technique which offers highest performance accuracy. The performance measures used are mostly standard type and hence the definition and formulae of these measures have not been provided.

7 Conclusion

A review of standard articles on CI based smart city community has been made and the important finding of the review has been presented in precise and systematic manner. It is expected that the researcher working in this field would find the paper interesting and useful. In essence, the paper has identified nine different directions of smart city in which development is taking place and research work is being carried out. It is, in general, observed that major investigations have been made in the area of intelligent transportation and security issues related to smart city. Less emphasis has been made in the area of smart education. Due to severity of COVID-19 in the whole globe, e-learning and smart education have become more important and hence more emphasis should be given for developing tools, techniques and platform to strengthen smart education system using CI. It is further observed that CI and particularly DL have become promising and essential methods for smart city community. Many more efforts need to be made to address and solve various other challenging problems related to smart city using ML and CI techniques. Very few of the proposed methods have been implemented in practice. Work can be carried out to develop prototype working online models to be used for the service of smart city community.

References

O’Dwyer

, Pan

, Acha

and Shah

, Smart energy systems for sustainable smart cities: Current developments, trends and future directions, Applied Energy 237 (2019), 581–597.

Ullah

, Al-Turjman

, Mostarda

and Gagliardi

, Applications of Artificial Intelligence and Machine learning in smart cities, Computer Communications, (2020).

Chen

, Wang

, Wu

, De

, Wang

, Zhang

and Huang

, A survey on an emerging area: Deep learning for smart city data, IEEE Transactions on Emerging Topics in Computational Intelligence 3(5) (2019), 392–410.

Bhattacharya

, Somayaji

S.R.K.

, Gadekallu

T.R.

, Alazab

and Maddikunta

P.K.R.

, A review on deep learning for future smart cities, Internet Technology Letters, (2020), e187.

Zhu

, Yu

F.R.

, Wang

, Ning

and Tang

, Big data analytics in intelligent transportation systems: A survey, IEEE Transactions on Intelligent Transportation Systems 20(1) (2018), 383–398.

Serban

A.C.

and Lytras

M.D.

, Artificial Intelligence for Smart Renewable Energy Sector in Europe—Smart Energy Infrastructures for Next Generation Smart Cities, IEEE Access 8 (2020), 77364–77377.

Zhao

, Chen

, Wu

, Chen

P.C.

and Liu

, LSTM network: a deep learning approach for short-term traffic forecast, IET Intelligent Transport Systems 11(2) (2017), 68–75.

Shi

and Abdel-Aty

, Big data applications in real-time traffic operation and safety monitoring and improvement on urban expressways, Transportation Research Part C: Emerging Technologies 58 (2015), 380–394.

Ning

, Dong

, Wang

, Rodrigues

J.J.

and Xia

, Deep reinforcement learning for vehicular edge computing: An intelligent offloading system, ACM Transactions on Intelligent Systems and Technology (TIST) 10(6) (2019), 1–24.

10.

Huang

, Song

, Hong

and Xie

, Deep architecture for traffic flow prediction: deep belief networks with multitask learning, IEEE Transactions on Intelligent Transportation Systems 15(5) (2014), 2191–2201.

11.

Hassan

S.U.

, Shabbir

, Iqbal

, Said

, Kamiran

, Nawaz

and Saif

, Leveraging deep learning and SNA approaches for smart city policing in the developing world, International Journal of Information Management, (2019), 102045.

12.

, Deng

, Gupta

B.B.

, Wang

and Choi

, A novel CNN based security guaranteed image watermarking generation scenario for smart city applications, Information Sciences 479 (2019), 432–447.

13.

Qureshi

K.N.

, Ahmad

, Piccialli

, Casolla

and Jeon

, Nature-inspired algorithm-based secure data dissemination framework for smart city networks, Neural Computing and Applications, (2020), 1–20.

14.

Shafiq

, Tian

, Sun

, Du

and Guizani

, Selection of effective machine learning algorithm and Bot-IoT attacks traffic identification for internet of things in smart city, Future Generation Computer Systems 107 (2020), 433–442.

15.

Chaouachi

, Kamel

R.M.

, Andoulsi

and Nagasaka

, Multiobjective intelligent energy management for a microgrid, IEEE Transactions on Industrial Electronics 60(4) (2012), 1688–1699.

16.

Azad

H.B.

, Mekhilef

and Ganapathy

V.G.

, Long-term wind speed forecasting and general pattern recognition using neural networks, IEEE Transactions on Sustainable Energy 5(2) (2014), 546–553.

17.

Mocanu

, Nguyen

P.H.

, Gibescu

and Kling

W.L.

, Deep learning for estimating building energy consumption, Sustainable Energy, Grids and Networks 6 (2016), 91–99.

18.

Zhang

, Liu

C.H.

, Tang

, Xu

, Ma

and Wang

, Learning-based energy- efficient data collection by unmanned vehicles in smart cities, IEEE Transactions on Industrial Informatics 14(4) (2017), 1666–1676.

19.

Mnih

, Kavukcuoglu

, Silver

, Rusu

A.A.

, Veness

, Bellemare

M.G.

,... and Petersen

, Human-level control through deep reinforcement learning, Nature 518(7540) (2015), 529–533.

20.

Sivanathan

, Gharakheili

H.H.

, Loi

, Radford

, Wijenayake

, Vishwanath

and Sivaraman

, Classifying IoT devices in smart environments using network traffic characteristics, IEEE Transactions on Mobile Computing 18(8) (2018), 1745–1759.

21.

Lopez-Martin

, Carro

, Sanchez-Esguevillas

and Lloret

, Network traffic classifier with convolutional and recurrent neural networks for Internet of Things, IEEE Access 5 (2017), 18042–18050.

22.

Gupta

S.K.

, Ashwin

T.S.

and Guddeti

R.M.R.

, Students’ affective content analysis in smart classroom environment using deep learning techniques, Multimedia Tools and Applications 78(18) (2019), 25321–25348.

23.

Hassan

S.U.

, Waheed

, Aljohani

N.R.

, Ali

, Ventura

and Herrera

, Virtual learning environment to predict withdrawal by leveraging deep learning, International Journal of Intelligent Systems 34(8) (2019), 1935–1952.

24.

Delgarm

, Sajadi

and Delgarm

, Multi-objective optimization of building energy performance and indoor thermal comfort: A new method using artificial bee colony (ABC), Energy and Buildings 131 (2016), 42–53.

25.

Logesh

, Subramaniyaswamy

, Vijayakumar

, Gao

X.Z.

and Indragandhi

, A hybrid quantum-induced swarm intelligence clustering for the urban trip recommendation in smart city, Future Generation Computer Systems 83 (2018), 653–673.

26.

Hernafi

, Ahmed

M.B.

and Bouhorma

, ACO and PSO algorithms for developing a new communication model for VANET applications in smart cities, Wireless Personal Communications 96(2) (2017), 2039–2075.

27.

Ali

, El-Sappagh

, Islam

S.R.

, Kwak

, Ali

, Imran

and Kwak

K.S.

, A smart healthcare monitoring system for heart disease prediction based on ensemble deep learning and feature fusion, Information Fusion 63 (2020), 208–222.

28.

Muhammad

, Khan

, Del Ser

, de Albuquerque

V.H.C.

, Deep learning for multigrade brain tumor classification in smart healthcare systems: A prospective survey, IEEE Transactions on Neural Networks and Learning Systems, (2020).

Comprehensive review of computational intelligence based smart city community

Abstract

Keywords

1 Introduction

2 Research gap, motivation of research, research objectives and organization

2.1 Research gap

2.2 Motivation of research

2.3 Research objectives

2.4 Organization

3 Literature review

5.2 Random forest (RF)

5.3 Deep reinforcement learning (DRL)

5.4 Deep belief network (DBN)

5.5 Multiple Regressions (MR)

5.6 Deep learning (DL)

5.7 Convolutional neural network (CNN)

5.8 Dragonfly algorithm (DA)

5.9 Moth-flame optimization (MFO)

5.10 Ant colony optimization (ACO)

5.11 Bayesian network (BN)

5.12 C4.5 algorithm

5.13 Naive bayes (NB)

5.14 Random tree

5.15 ANN Ensemble (ANNE)

5.16 Fuzzy expert system (FES)

5.17 Multi-objective optimization (MOO)

5.18 Recurrent neural network (RNN)

5.19 Support vector machine (SVM)

5.20 Conditional restricted boltzmann machine (CRBM)

5.21 Factored conditional restricted boltzmann machine (FCRBM)

5.22 Artificial bee colony (ABC) algorithm

5.23 Quantum induced swarm intelligence (QPSO)

5.24 Particle Swarm Optimization (PSO)

5.25 Feature fusion

6 Analysis of results presented in various articles

References

5.12 C_4.5 algorithm