Abstract
The Severe acute respiratory syndrome coronavirus (SARS-CoV) are deadly infectious disease which can easily transmit and causes severe problems in humans. It is known as a coronavirus and referred as a common form of virus that naturally causes upper-respiratory tract illnesses and the symptoms are hard to identify. It is important to recognize the patient and providing them with suitable action with constant intensive care. Healthcare amenities is constructed on fog and big-data based system and it is integrated with cyber-physical system. The role of Cyber physical system in health care domain is to fetch deep insights about the nature of disease and carry the monitoring process with early detection of infected users. The objective is to identify occurrence of SARS at initial stage. In proposed system, resemblance factor is evaluated from the extracted keywords. In order to identify the difference between SARS affected and others, the proposed scheme fetches the inputs from user’s displayed in the form of text. It is passed to deep recurrent neural network (RNN) model. It extracts useful information from the raw information given by the user. The J48graft algorithm is used to carry the classification based on the type of infection and symptoms of each user. The data is stored in the bigdata layer (mongoDB) and it detects the infected area by using the geospatial feature in mongo dB. The methodology is framed in the proposed model to prevent the spread of disease to other users. In case of any abnormality the generation of alert process is done instantaneously and directed on user’s mobile from fog layer. The final experimental outcome reveals information about the performance of proposed system in terms of Success rate, failure rate, latency and accuracy %. It shows that the proposed algorithm gives high level of accuracy when it is compared with other primitive methods.
Keywords
Introduction
The severe acute respiratory syndrome coronavirus (SARS-CoV) was an extensive hazard allied with substantial sickness and death in the initial 2000s. The SARS-CoV produce brought 8,096 cases, including 774 deaths, in 37 countries within 8 months [10, 19]. After many researches, the researchers found out the outbreak by understanding the characteristics of SARS-CoV and its occurrence of spreading in wild animals. Comparatively several other respirational viruses, the SARS-CoV is alleged from infested person to the vulnerable [16]. It transforms through routes, i.e., far-off aerial, adjacent connection and which can easily carry or transmit the disease in different routes. Fetching the information about the nature of the diseases and getting deep insights about the spreading medium of the disease was considered to be a complex task. Sometimes the researchers should know about the control measures such as a) standard, b) contact, c) droplet and d) precautions for airborne. These factors don‘t have clear explanation about the handling of disease once it gets started. The objective of the work is carried by adopting the upcoming domains such as fog computing, cyber physical system and deep recurrent neural network (RNN). The contribution of these domains helps to find the nature of the disease and it also identify the countermeasures for affected person by creating a boundary and separating healthy people from the infected person [1]. The fog computing achieves instantaneous examination of health based complex data produced by medical policies [6]. The fog computing has the ability to analyze the data with low latency, and it guarantee quality of service with instantaneous notification to healthcare applications. The proposed system compresses of data gathering phase from the variety of users and devices. The details such as personal information about the users are collected in the form of text messages either from web browser or mobile applications. The role of cyber phase is to extract crystal data for carrying analyzing process. The cyber phase contains a fog layer which helps to identify and extract needed keywords by adopting a method known as deep recurrent neural network (RNN). The model is used to combine keywords and context information [18]. It contains a symbol in front of the system and each symbol gives the information about the level of severity. For example, let us assume # headache / $stomachache are the keywords to be noted in a big phrase. This key phrase can easily construct or extract the information needed for identifying the information about the user.
Flow of general system
The message is identified as general information shared by the affected person with more details. The proposed model helps to extract crystal information to find the appropriate words which gives a proper key words for further processing.
For example: A person mentions the name with the symbol @ and other sort of information is followed by a symbol like $ their locality, address and state etc. The # symbol is used for representing the name of the disease. The proposed method extracts the key information from the sentence and gives crisp data for carrying effective processing of data. The example gives the ideology of the key extraction process in the model.
Hello, My name is
According to the data symbols were derived with its severity, and assume some rules for easy processing like the user name should be kept with symbol
First, filtering out all unwanted statements or fillers in the sentence is done, then removal of unwanted URL links is carried because the proposed method mainly focusses on the key details and not any unrelated information. Moreover, the system is designed with some rules about the symbols (@- username, $ location details, # for symptoms and disease name). The disease should have a separate # hashtag symbol and each disease or symptoms should definitely have their symbols in the front of the starting word, which is considered as key phrase. Every key phrase is mapped with a factor or score. And the key phrase is compared with SARS domain wordlist. It helps to differentiate symptoms-based SARS diseases using similarity measures. The classification of affected person is done by implementing classification technique. In this paper adopting Decision tree for executing the classification process is done. The user’s cluster mapped into uninfected, infected, vulnerable of corresponding disease.
The novelty of the proposed is organized in this paper are: To design a model for identifying the occurrence of SARSs at an initial stage by carrying the key extraction process using RNN. To adopt a J48graft algorithm for carrying the classification process. The design uses mongo DB for storing the data and it has the ability to detect the infected area by using the geospatial feature. This methodology is mounted in the proposed model in order to avoid the blowout of disease to other users.
To achieve the objectives, Bigdata fog is constructed with cyber physical system for carrying the monitoring and analyzing the effectiveness of disease in humans.
The forthcoming section of the paper is systematized as trails. Section 2 defines literature study of the cyber-physical based health care system in feature. Section 3 defines the characteristics of proposed model and the Section 4 illustrates experimental outcomes and performance examination of the proposed system. Lastly, Section 5 describe completeness of the paper.
Literature study
Zhang, Yin, et al proposes a framework using the concept of Cloud and Big Data which recommends a cyber based physical system for victim. It also added with health management submissions and facilities, called Health-CPS [17]. It was constructed using cloud computing with big data analytics for carrying the process of data analytics. The system contains data collection also known as data gathering layer with an integrated standard, organization of data for dispersed storage. The data-oriented facility layer describes about the service of layers based on the data. The outcomes of the learning demonstrate the knowledge of cloud based big data model for improving the presentation of the health management organization. Beliga, Slobodan, Ana Meštrović, and Sanda Martinčić-Ipšić analyses approach and tactics for the mission of extraction of keyword. The organized analysis of methods was collected and gives complete examination of traditional approaches [2]. The keyword extraction was expanded and it uses supervised and unsupervised methods. It also explains about the special emphasis on graph-based method and methodology. The graphs were analyzed and compared to fetch deep insights and guidelines for carrying next step in the research. The expansion of new graph procedures for keyword extraction also used in algorithm for carrying efficiency. Mahmud, Redowan et.al. proposes a Fog-based IoT-Healthcare description with proper design and discover the mixing of Cloud-Fog facilities in interoperable Healthcare explanations stretched upon old Cloud-based design [12]. The situations were presented in the paper for carrying the evaluation process. iFog Sim simulator is used for carrying the simulations. The outcomes were examined in relation to dispersed computing, latency was minimized, optimization process was done when the data gets communicated, with less consumption of power. The experimental outcomes show enhancement with less cost, minimized delay of network and less consumption of energy. Dogaru, Delia Ioana, and Ioan Dumitrache displays the reputation of cyber-physical systems in healthcare systems by suggesting an overall framework or model for the organized medical based devices [5]. It reveals problem faced by the system while carrying service. The theme of the paper was described the loopholes in the healthcare schemes and discloses the lack of security in the medical part. Basco, J. Antony, and N. C. Senthilkumar illustrates the technology with Big data analytics. BDA in to Healthcare division and Health Big Data Analytics (HBDA) will deal with complex patient and the detail of those affected person mostly in unstructured arrangement of data. The data was collected from different inputs and analyzed effectively [3]. The data was fetched from the prescriptions, data retrieved from images and reports. The problem of analyzing more sets of data was difficult because of the size and variety of medical data. Here the author produces Electronic Medical Records (EMR) formed from several medical strategies and portable applications will be brought into MongoDB using Hadoop framework with Improvised processing technique to expand result of processing patient histories.
The literature review explains the ideology of proposed system with details of research gaps are displayed in Table 1. It reveals the theme of the evolving of proposed system by overcoming the issues of other traditional approaches. The demerits of the primitive methods are considered as a key point and according to it the framework of proposed model is generated.
Related work of sars and domain based health care system
Related work of sars and domain based health care system
The proposed scheme is displayed in Fig. 1 compress of two key phases such as physical phase and cyber phase. The role of physical phase is to gather all type of information from the users such as personal information, environment-based information, medical issues. The cyber phase uses fog layer, for transmitting acquired data, to carry real time processing and to identify the infected and uninfected users in the environment [7]. The Tables 2–4 represents the information required for proposed system.
Complete design of the proposed system.
User personal details
User personal details
Symptoms and keywords
Symptoms based comparison of SARS diseases
The role of cyber phase is to predict and analyze the health statistics gathered from the physical phase
Algorithm: 1) Keypharse extraction algorithm from the user generated query and 2) Mapping of extracted query with SARS
and it is mainly used to fill the gap between the communication of messages. It shows actual time alert messages to user about the severity of infection of SARS [6]. It is connected to the big data phase for storing purpose and compared with thesaurus.
Further it is distinguished and classified into keywords for effective purpose. The DRNN collects personal information about the users and their medical records, symptoms based on SARS disease. It carries the process known as extraction of keywords for converting the unclear text data into crisp in useful data format. The process supports the system to understand the impact of disease in the user and it further aids to give countermeasures by distinguishing the infected and uninfected users. The DRNN is used for extracting useful information by removing unwanted data and the remaining words gives the meaning and level of severity of the disease. The contender keywords are identified which can be used for future analysis purpose [11]. After finding of contender keywords occurrence and frequency of individual word is evaluated. The occurrence of words defined as the sum of words occur in text and precision of word (word looks in the document). The word Score (WS) is evaluated using the equation 1.
After evaluating the word score of individual contender word, rank selection method is used. The individual’s words in the key phrase which have very close word score is nominated as a keyword fetched from the user’s message. The mined keywords are denoted in the arrangement keyword list. The keyword list compresses of sum of keywords are produced by the DRNN algorithm [4]. It also reveals the symptoms, location, treatments, causes, ecological and personal information and the keywords are categorized in the form of group represented in the tables. The symptoms like body ache, joint pain, fever is grouped into single category. And other keywords are placed to their corresponding category.
The domain analogue (thesaurus) is built depends upon the SARS disease data, the role of domain analogue is to support keyword extraction. The users enter their own text, based on their environment condition. So, the domain analogue identifies the words and it is compared with analogue, to extract the perfect keywords for processing. For example, if a user enters the term arthritis as an input and other user enters joint inflammation.
It indicates similar meaning and according to the domain analogue it provides similar information and it is converted into standard words such as Joint pain, which can be easily classified for effective classification. Once the user text is converted into a proper format. The distinguish and classification of user’s symptoms placed under the three SARS major symptoms [4, 10]. The Table 4 displays Symptoms Based Comparison of SARS Diseases with the severity level.
The resemblance aspect ratio between users and three Major SARS. And the severity level is placed in the form of *** for high level of severity major symptoms where the users who falls under this category contains all sort of symptoms of SARS, ** moderate severity level not so significant but still noteworthy indicators of these viruses, * no issues. For identification the implementing of evaluation method such as similarity factor, resemblance aspect ratio, probability allocated for each symptom is done. Here, ‘***’ remains allocated the rate of ‘1’,
The symbol representation of ‘**’ is allocated the value of ‘0.5’, ‘*’ is allocated the value of ‘0’.
Where ‘U‘ denotes the user and ‘S’ indicates the SARS i.e. 3 major Viral pneumonia, dyspnea, Respiratory failure (pneumonia, respiratory failure, heart failure, and liver failure, fever, joint pain).
Here RAR(U,S) denotes the Resemblance Aspect Ratio between user ‘U‘ in equation 3 and SARS ‘S’ in equation 2.
The decision trees play a vital role in the classification and decision making for a problem. The key features such as its in-memory classification models, less computation costs, and no need of maintaining frequent database lookups [8]. It has the capability to handle a dataset with a high degree of errors and misplaced values. It doesn‘t make assumptions about spatial distribution or the classifier‘s structure [1]. Random Tree (RT) is an effective algorithm for building a tree with K random features. It can be generated efficiently and the combination of large sets of random trees generally leads to accurate models. The random tree can easily fall to overfitting (it will give outcomes with high amount of error). Reduced Error Pruning (REP) Tree is the simplest and most logical method in decision tree pruning. It is a quick decision tree learner, which generates a decision tree with the help of information gain. It also used for carrying the splitting condition and trims it for achieving less error pruning. J48graft algorithm helps to enhance the probability of accurately categorizing the instances. The algorithm produces only solo tree which is mainly done to minimize the error of prediction. J48 tree algorithm helps to derive the J48 graft. The grafting procedure improvise possibility of categorizing instances placed external with the parts enclosed by training data [9]. The purpose of using J48graft algorithm is to enhance the probability of classifying the data in an accurate way and to minimize the error. J48 algorithm is used as normal classification technique for generating grafted decision tree. The advantage of using J48graft is it can be executed within less amount of time. The persistence of this grafting algorithm is to enhance accurate classifying of data from the raw data generated by the users. The J48 grafting algorithm provides the finest overall prediction accuracy over an illustrative choice of the learning process.
The imbedding process increases node to conclude decision trees with persistence of minimizing error prediction [12]. The J48 grafting procedure delivers finest wide-ranging prediction with an accuracy over a typical choice of the learning process. The generation of J48graft using WEKA and it is characterized as UNI if the user U has no infection, but having symptoms like Conjunctivitis (CJ), Immune thrombocytopenic purpura (ITP) and Stomach cramps (SC). However, if a user has CJ, high fever, SC and Muscle Pain (MP), type of contamination will be pushed in the category of IN [20]. The manipulator takes a huge volume of infection can be transported to further VL users or persons. The user is measured as UN if they do not have any of mentioned health indications.
Mongo DB
The variety and value of data in medical field is massive and in the cyber physical system the data is generated from variety of IOT devices for measuring and processing [14]. The advantage of using MongoDB in healthcare domain is to provide flexible data model using powerful language query. The nature of MongoDB supports the user by providing effective analytics and security. The type of fields values in MongoDB mainly uses Geo-Coordinates which services to identify the infected user’s area and it help to separate the uninfected users. The sharding logic is a salient feature in MongoDB for distributing the files to multiple users, these feature shares the information to healthcare center, hospitals etc. The location of the user is collected from query and it is passed to mongo DB. The mongoDB has GeoJSON objects to process the location-based information and it can calculate the distance between the users. Here, in our proposed model a storage and processing layer is adopted. The Mongo DB is used for storing and processing the queries. It also contains a feature, of encrypting the data in order to provide security. The protection of data should be done effectively by adopting an integral feature known as geospatial within the database of MongoDB’s storage of encrypted engine. It also compresses of a key feature known as geo-capabilities, it can store GeoJSON objects and the geospatial indexing applied. It has become progressively significant for health based and disease investigations in identification and classification of disease for carrying analyzation.
Let’s use a standard detail displayed by the user as an example. The message may contain geographic information a data which denoted location-based information as coordinates approaching from either the user’s GPS enabled smartphone or their system location settings. And the MongoDB uses geo-capabilities and it can store GeoJSON objects, and has geospatial indexing implemented in it. The message can be secured. The process of encryption is enabled, which uses Advanced Encryption Standard with 256 bits embedded with Cipher block chain. AES-256 is known as a convolution key; i.e. the same key is adopted for carrying both encryption and decryption of text [13]. Mongo DB helps to store the entire information in a single place for effective storage and less time for processing. The major advantage of using mongo dB is early stage of identification of disease, analyzing the information based on the keywords in an efficient manner. The authentication and authorization are the inbuilt functionalities of using MongoDB. The closeness information among users is also stored this layer, which helps to create and update information for carrying analytics using mongo DB.
The analytics gives us some information for understanding the spread of the disease among the user. The mongoDB compass is used for analyzing and visualization the data. It defines the nearness of any user with Uninfested (UNI), Infested (IN), Vulnerable (VL) [8]. The equation 4 and 5 describes the information about the nearness and extension ratio. It is premeditated as the inversely proportionate to summation of sequential less distance from user ‘a‘ to user ‘b‘. The expansion ratio defines fastness of SARSs which blowout from an infested user, to another user. If the extension ratio is less then possibility of quick disease transmissions to other users.
Experimental setup and examination of performance are classified such as initiation of data, Resemblance aspect ratio estimation, effectiveness of J48 graft decision tree, estimation of mongo DB based outbreak metrics. The metrics used for illustrating the performance of the J48Graft with other models such as REPTree (RPT), and random tree (RT) is denoted as 1) Accuracy, 2) Success Rate and 3) Failure rate. The latency of MongoDB is calculated and it is compared with Hbase, Cassandra.
Initiation of data
An illustration query correlated to virus is submitted by john and it is used for validating text file. The mentioned query is used to demonstrate the proposed model, where the query is submitted to the system. The deep recurrent neural network (RNN) is used to combine keywords and context information [2]. It contains a symbol in front of the system and each symbol gives the information about the level of severity. The Figs. 2 and 3 defines an example such that: # headache / $stomachache are the keywords to be noted in a big phrase. Based on this keyphrase the construction or extraction of the data for further analyzing is carried effectively and it is displayed in Figs. 2 and 3.
User information in the form of queries.
User information after adopting RDNN algorithm (Keyword Extraction).
The deep recurrent neural network (RNN) is functional to 1058 queries to produce appropriate keywords [16]. These keywords are related with SARS domain analogue, and it is used to classify based on the symptoms of users, personal information, medication taken by the user etc. The three major SARS diseases such as fever, joint pain, Respiratory failure are considered for calculating Resemblance Aspect Ratio for symptoms expressed by query of each users. The Resemblance Aspect Ratio is executed and identifies 598 queries of respiratory failure, 200 of joint pain, 260 fever [17]. Resemblance Aspect Ratio of the proposed scheme are related with two additional approaches known as coefficient of correlation and regression to compute similarity factor illustrated in Table 5. The list of Keyword is produced by DRNN is given as input to evaluate the performance of these three methods. The Table 5 illustrates the information about the performance factors such as accuracy, specificity, precision, recall, relative absolute error and root mean square error. The proposed scheme gives accuracy of 90.5 by adopting J48 graft classifier displayed in Fig. 4. The role of the J48 classifier is to enhance the correctness of the SARS disease with less amount of error rate (Relative absolute and Root mean square error). The statistical parameters like precision is mainly used for evaluating the exactness of the classifier and the value of precision is 90.4 which illustrates the efficiency of classifier has the ability to classify the SARS disease among user. The parameters help to gain information about the proposed scheme with 5 main metrics such as accuracy, level of precision and recall with errors (RMSE and RAE). The values indicate that proposed scheme gives less amount of error with high level of exactness and accuracy for identifying the SARS disease among users.
Comparative results for resemblance ratio
Comparative results for resemblance ratio
Accuracy (%).
J48graft classifier carries the extra classification process. The classifier helps to categorize the based on the query like Uninfested (UNI), Infested (IN), Vulnerable (VL). A decision tree is executed in Weka 3.6 to estimate the presentation of J48graft decision tree. Used for representation, J48graft decision tree for all three diseases. In Weka 3.6, 1058 queries are scrutinized and several numerical procedures were processed to understand the efficiency of algorithm. Success rate indicates to the percentage of SARS categories, which can be precisely classified by the classifier. Failure rate denotes to the percentage of SARSs which are classified inaccurately. The Figs. 5, 6 displays J48graft provides a lower FR (Failure Rate) and higher SR(Success Rate). Precision and recall are arithmetical procedures for correctness and wholeness of classifiers. If the classifier gives maximum value of both recall and accuracy, then automatically the precision of will moves towards high percentage. J48 decision tree provides maximum recall value and exactness displayed in Figs. 4–6 respectively. Hence, the arithmetical outcomes display the proposed scheme with J48 decision tree is proficiently categorize the group of each user with maximum level of classification accuracy, sensitivity, and small specificity value with less amount of errors. The classification algorithms such as REPTree (RPT), and random tree (RT)is adopted for comparing process with proposed method.
Failure rate (%).
Success rate (%).
The theme of using mongoDB is to store and analyzing the data retrieved from the physical phase and J48gaft.The mongoDB collects information about the infected users which is classified. It helps to identify the users exposed to the disease and gathers all sorts of information for further analyzing. MongoDB geospatial helps to identify location or area of the disease with the help of Deep recurrent neural network. The Latency is calculated by finding the average amount of time acquired by the MongDB server to regain and return the data from the database.
Latency rate is the key factors used for evaluating MongoDb and it is compared with other approaches such as HBase and Casandra. The Fig. 7 depicts the evaluated delay related to time for mongoDB, Hbase and Casandra. The location of the user is collected from query and it is passed to mongo DB. The mongoDB has GeoJSON objects to process the location-based information and it can calculate the distance between the users. The time taken for reading and writing the query using mongoDB is less when compared with another database. This is because of the fast query processing the latency of the database for processing the query is less. The Cassandra and HBase are other two database used for carrying the comparison with MongoDB. The latency of these two databases are high because of its query processing technique.
Latency vs time.
In MongoDB the complex data can be effectively processed because the query is modeled using JSON format. The workload (query request and response) in these evaluations illustrates that MongoDB provides less latency than HBase and Cassandra. The results display the latency rate for storing and analyzing the user infection category is less compared with HBase and Cassandra. This is because of the key features in the mongoDB. It uses MongDB Geospatial, to gather all information of the infected users for analyzing. It can easily classify the difference between infected and uninfected user. The number of data processing using MongoDB consumes less latency by giving the quick and effective results. Hbase and Cassandra don‘t have any key features for analyzing the difference between the infected and uninfected users and it will take high amount of time to process the query generated by the user which leads to consume delay. The MongoDb compress helps to create a graph with detailed visualizations in Fig. 7.
SARS are disease leads the user to major problems and create a great impact in the society. The rapid growth of the current technology such as IOT sensors, smart devices helps to provide solution by adopting the fog layer with cyber physical system for classifying and gathering of data from different users.
The bigdata analytics method mainly used for storage and processing the data retrieved from the raw data. In this paper the fog and bigdata based cyber physical system is used for effectively identifying the disease in the users. The user information is derived in the form of query. The j48Graft is used for carrying the classification process between the infected and uninfected user based on the severity level. The factors like accuracy and success rate of the proposed scheme is evaluated and the results of the proposed method is high when related with other traditional approaches. The Recurrent Neural Network method help to extract keywords from the user queries. It also contains a storage layer for collecting all sorts of information for further analyzing. MongoDb geospatial helps to identify location or area of the disease by the help of Deep recurrent neural network. The results generated by the methods are evaluated by the key performance metrics and the results are compared with other traditional methods. The future work of this system is to enhance the classification scheme to improvise the segregation of infected and uninfected user. In this paper text-based input from the users is given as input for processing the data and our future work will depend upon other sets of formats such as image, video and audio.