An intelligent parking allocation framework for digital society 5.0

Abstract

In recent smart city innovations, parking lot location has garnered a lot of focus. The issue of where to put cars has been the subject of a lot of literature. However, these efforts rely heavily on algorithms built on centralized servers using historical data as their basis. In this study, we propose a smart parking allocation system by fusing k-NN, decision trees, and random forests with the boosting techniques Adaboost and Catboost. Implementing the recommended intelligent parking distribution technique in Smart Society 5.0 offers promise as a practical means of handling parking in contemporary urban settings. Users will be given parking spots in accordance with their preferences and present locations as recorded in a centralized database using the proposed system’s hybrid algorithms. The evaluation of performance considers the effectiveness of both the ML classifier and the boosting technique, and it finds that the combination of Random Forest and Adaboost achieves 98% accuracy. Users and operators alike can benefit from the suggested method’s optimised parking allocation and pricing structure, which in turn provides more convenient and efficient parking options.

Keywords

Parking space administration machine learning control scheme hybrid-mechanism k-nearest neighbour

1. Introduction

One of the most significant uses of smart technologies is in parking space management, which has revolutionized the way in which people interact with everyday goods. Parking availability, occupancy, and duration data are all gathered in real time by Internet of Things (IoT) sensors and cameras in smart parking management systems. As a result of this analysis, motorists are provided with up-to-date parking place information, and parking lot managers are able to streamline their services [1, 34].

Using real-time data, intelligent parking management systems dynamically allot parking spots and adjust parking rates to optimize space use and revenue. Parking demand can be estimated and space allocation optimized with this information, leading to streamlined operations and reduced traffic jams. Intelligent parking systems (IPS) enhance user satisfaction by facilitating cashless payments, real-time parking availability, and guidance through the use of mobile apps or digital signage. The advantages of an intelligent parking system are shown in Fig. 1 [5, 31].

Figure 1.

Advantages of IPS in digital soceity.

In addition to enhancing effectiveness and accessibility, sophisticated parking administration techniques can lessen their negative effects on biodiversity. Congestion in the public transportation system is alleviated, and air quality is enhanced, thanks to these initiatives. When utilized alongside shared bikes and public transportation, efficient parking facilities can help develop more environmentally friendly urban mobility ecosystems [4, 8].

Intelligent parking administration systems, as depicted in Fig. 2, are notoriously challenging to put into practice [3, 16]. Data security and confidentiality are crucial concerns. These systems need to have acceptable security requirements to capture vital parking data without worrying about data breaches or illegal access. Compatibility issues can also make it difficult to use many smart parking systems and gadgets together [8, 9, 10, 11].

Figure 2.

Concerns about computerized parking implementations.

Smart parking management systems might be expensive to implement. These systems’ initial investment can be high, but they can generate long-term savings and revenue generation options. That’s why it’s so important to weigh the costs and benefits thoroughly before getting started [18, 32]. Modern parking management systems [8, 33] are a step toward a more eco-friendly and productive future. Integrating with various green transportation approaches, gathering and analyzing information in real time, and implementing dynamic rates and parking space allocation are all ways in which these technologies may help customers and parking managers. Effective implementation of such platforms requires thought to be given to issues of security and confidentiality of information, interoperability, and cost [10, 12].

In this effort, modern methods are used to enhance the smart parking management system. An adaptively enhanced random forest method is offered to facilitate user choice of parking locations. To better predict parking slot selections, the computer learns from user behavior and preferences. A more efficient and individualized user experience is produced. The second section of the research provides recommendations for where to put parking permits in the future to keep reserved spots from being taken away. The system is able to determine when a user no longer requires a designated area by monitoring their activities and the flow of traffic. Reduce the number of vacant parking spots by giving one to another driver. In the latter section of the paper, the need for collecting and managing data in real time is emphasized, both for database entries and experimental datasets. Because of this, the system can easily adjust to new circumstances by analyzing user behavior and system health with great accuracy and speed.

Here’s how the rest of the tasks are laid out: The literature on parking administration is dissected in the following part. In the third section, we’ll talk about collecting features, hybrid ML, and evaluation metrics. Many studies and experiments are discussed in the fourth part. The final section wraps up the whole work.

2. Literature study

Data shows an increase in both traffic congestion and wasted time in the suburbs of major cities. This is a waste of gas, an annoyance to motorists, and an environmental hazard [39]. Congestion has been shown to have an impact on gas consumption [5, 21]. As a result, there is an increase in the discharge of NOx, CO, CO2, VOCs, HCs, and VOCs. According to the United Nations Environment Programme [17], air pollution is responsible for about 7 million fatalities per year. Another study from the Harvard School of Public Health predicts that congestion will cost the economy $100 billion by the year 2020 [30].

Parking facility occupancy estimates are promoted through an emphasis on the capacity to resolve parking issues [5, 6, 7, 8]. By utilizing historical vacancy data, this demonstrates the process of generating such assumptions. The paper showcases experimental evaluation outcomes of different methodologies that utilize publicly available databases, well-established regression algorithms, and a meticulously curated collection of features that have been developed since the initial publications [4] and [38].

Scientists have created multiple SPS (Smart Power Systems) utilizing diverse techniques and sensors. In reference [3], the authors provided a concise overview of their original concepts and took into account the attributes of several modern computerized parking methods. The article covers the topics of SPS approaches, indicators, and interfacing equipment. The paper does not address the techniques or procedures that establish the structure of systems that directly interact with users. Nene et al. [34] introduced comparable car parking systems by employing intelligent parking technologies. The author suggests online voting as a means to ease traffic and minimize time in parking lots. The Hungarian distribution mechanism and online voting are demonstrated in [13] to effectively reduce parking shortages and enhance convenience.

The article [44] examines the use of AI and ML in the areas of energy management, public service, security, and protection in the context of smart cities. Congestion mitigation at traffic lights, off-peak energy use in buildings, improved trash management, and enhanced security can all be achieved through these applications. The author offers a creative approach utilizing artificial intelligence to keep an eye on traffic jams within the context of the Internet of Vehicles (IoV) [46]. In an effort to foresee and avoid traffic congestion, the article examines the difficulties associated with managing the increase in connected cars that contribute to vehicular congestion. The suggested approach consists of gathering real-time traffic information from vehicles in motion, applying machine learning algorithms such as Support Vector Machine (SVM), Decision Tree (DT), and Random Forest (RF), processing and analyzing the information, and subsequently generating predictions regarding traffic congestion [2, 3, 4, 5, 6].

Table 1
New developments in the field of intelligent parking systems

Year	Author	Inference
2017	Huange et al. [19]	The identification of parking spaces was accomplished by employing a multiple-layer interference architecture.
2019	Lin et al. [31]	An automated parking structure that anticipates the flow of parking-lot traffic and examines driving habits in order to maximize the use of available parking lots.
2019	Fulman et al. [15]	The parking spots are distributed using an adaptive spacing approach that takes into account the need and availability of parking spots for self-driving automobiles.
2018	Bura [3]	Hungarian allocation Model is employed for intelligent parking slot administration.
2019	Tu [39]	The regularity and predictability of varying parking and driving habits are evaluated.
2013	Chawathe [6]	The author provides a forecasting model that can improve parking management and cut down on traffic.
2020	Ye et al. [42]	Short-term parking spot forecasting is done using common machine learning methods.
2019	Weng et al. [41]	A gradient-boosting decision tree technique, among the various recurrent DT approaches to SPS, was put forward.

The authors in [30] provides a comprehensive analysis of the challenges, uses, and prospects of IoT wireless sensors. According to the authors, wireless sensor networks face challenges related to resource administration, safety, and data interpretation. In addition, they evaluate wireless sensors used in intelligent residences, medical care, and ambient surveillance IoT solutions. The experts conclude by providing their predictions regarding the utilization of wireless sensors in the Internet of Things (IoT) and emphasizing the imperative for more investigation. Table 1 is a compilation of current parking allocation systems.

These days, sensors and cameras are standard equipment for parking lot control. It’s effective, but it’s costly to keep up and put into practice. In order to locate open time slots, the author [29] employs the utilization of a recording device and computational imaging (YOLO); however, this method is both costly and picky about its parameters. Despite the necessity of accurate information, the vast majority of distributed systems merely offer empty spots in the garage. Having all of your database records up-to-date will speed up your job search. No sensors are required, so the cost of deployment is low. Sensor parking relies on a server to store data [21, 24]. Using the user’s historical and current data, the hybrid adaptively augmented random forest algorithm in the provided assignment model may make predictions about the user’s preferred location. Thus, cancellation due to prolonged slot retention is avoided.

3. Proposed methodology

3.1 Feature extraction

Intelligent parking systems (IPS) make use of machine learning (ML). Spot, availability, capacity, time-based trends, reservations, pricing, distance from attractions, ease of access, safety measures, current data, user reviews, and environmental impact are all extracted by ML. Allocation of resources, user satisfaction, and parking lot management are all enhanced by ML features [7]. With the use of ML, the system is able to locate and navigate parking spots, track availability, foresee occupancy, and dynamically assign spaces in response to past usage. ML algorithms are useful in reservation systems and pricing comparisons. Customer convenience is increased with the use of ML in the analysis of accessibility and attraction proximity [33]. Processing data in real time and incorporating feedback from users boosts decision-making and system efficiency. The system’s consideration of both green spaces and carbon emissions encourages sustainable parking [42]. Machine learning contributes to the development of smarter and more efficient parking systems.

3.2 Suggested IPS methodology

As illustrated in Fig. 3, the strategy that is advised is broken up into two stages, which are training and testing. Regularly revising a hybrid machine learning model using previously gathered parking data during the training stage results in an improved level of functionality. The machine learning algorithm is educated to analyze the collected data, including occupancy rates and trends in accessibility, in order to make a precise forecast of the availability of parking spots contemporaneously. The model’s ability to continuously evolve and adapt to ever-changing parking situations is enhanced by the use of past data, which also contributes to the model’s improved precision as well as efficacy.

The physical realm is optimal for doing system testing. Implementing real-time machine learning system assessment improves the effectiveness of parking demand handling. The Hungarian model selects the optimal parking location for a customer by considering factors such as proximity, cost, and accessibility [29, 33]. The server’s analytics section displays the allocated parking spaces for users. The user receives a parking ticket. By utilizing this token, users are able to authenticate parking upon their arrival. Upon token confirmation, the server stores and upgrades transaction information to retrain the system [43, 47].

Figure 3.

Proposed scheme.

The information that users provide will determine the effectiveness of the intelligent parking system. In order to train and fine-tune the model, data is used. With more information, the model can improve its predictions of parking availability and return more precise results to users. It is possible that the system will be able to anticipate changes in demand and adjust the distribution of parking resources accordingly.

The safety of the vehicles is paramount to the success of the IPS. The system in question must ensure the safety of both users’ personal information and equipment. The system employs cryptography and other safeguards to keep user information secure while in retention and transit. Parking lots and goods are protected from vandalism and burglary with the use of CCTV systems and other safeguards.

3.3 Estimation using a combination of ML model

Our suggested procedure makes use of a hybrid machine learning model [22, 23] that is routinely fed new parking records in order to improve its accuracy. In order to effectively anticipate parking spot accessibility in real time, this algorithm analyzes key parameters such as rate of vacancy, accessibility patterns, proximity among parking lots and locations, and charges. The model improves its accuracy and efficiency over time thanks to its ability to learn from its past mistakes and adjust to new parking situations.

Algorithm 1: K-NN with adaboosting
⟢ Inputs: Distance (D), Present request P, Transitional Forecast Q Outputs: Parking Label F
Step 1. for m $=$ 1 to I do:
Step 2. Set weights for all training inputs: $w_{i}=\frac{1}{\left\|{\textit{Distance}}\right\|}$
Step 3. for u $=$ 1 to T do:
Train k-NN on D with weights $w_{i}$ , and calculate Q for present P
Measure the error rate of k-NN, $\textit{err}_{t}$
Calculate the weight of k-NN, $\alpha_{t}$
Modify weights of training inputs: $w_{i,\textit{new}}$
Normalize weights: $w_{i}=\frac{w_{i}}{\textit{Sum of }(i)\textit{ weights}}$
end for u
Step 4. end for m
Step 5. Combine k-NN with weights $\alpha_{u}:F(P)$
Step 6. Predict class labels for $P$ using $F(p)$ and verify the evaluation measures if not satisfies go to Step 2.
Step 7. Allocate best F to the current request P

AdaBoost, also known as Adaptive Boosting, is a technique that has been modified for the purpose of classifying jobs [22, 25]. Multiple k-NN models are built on the a database, with individual weights assigned to every instance. Each sample commences with uniform weights. The dataset is used to train k-NN models with these weights, and their error rates are calculated. The weight of the k-NN model is determined by calculating it utilizing the error rate and the learning rate variable ( $\alpha$ ). The training sample parameters are adjusted based on the outcome of the k-NN model. The k-NN model assigns larger weights to misclassified data and lower weights to well classified samples. The weights are normalized to a value of one.

Algorithm 2: K-NN with catboosting
⟢ Inputs: Distance (D), Present request A, Intermediate Prediction B, $\gamma=$ 0.5, Prior predictions X, weight of prior value $\beta=$ 0.2 Outputs: Parking Label F
Step 1. for i $=$ 1 to J do:
Step 2. Initialize weights for all training samples: $w_{i}=\frac{1}{\left\|{\textit{Distance}}\right\|}$
Step 3. for t $=$ 1 to U do:
Train k-NN on D with weights $w_{i}$ , and predict B for current A
Estimate the error rate of k-NN, $\textit{err}_{t}$
Compute the weight of k-NN, $\alpha_{t}$
Update weights of training samples: $w_{i,\textit{new}}$
Regulate weights: $w_{i}=\frac{w_{i}}{\textit{Sum of all weights}}$
end for t
Step 4. end for i
Step 5. Combine k-NN with weights $\alpha_{t}:F(A)$
Step 6. Train CatBoost with $F(a)$ as the target variable $F(a)$
Step 7. For the predict class label $F(a)$ verify the evaluation measures if not satisfies go to Step 2.
Step 8. Allocate best F to the current request A

In order to enhance prediction, this boosting technique employs k-NN and CatBoost. The initial value of the weights used in training is 1/n. The k-NN approach employs weights w_i to train on data set D. Both the error rate and the learning rate are utilized to determine the k-NN model’s relative importance. Adjust the weights in the training model to account for the k-NN model’s error rate. This occurs t times out of t. This method combines the power of k-NN with that of CatBoost. Weights are used to train k-NN on the dataset, with the weights being determined, updated based on the training samples, and normalized. The CatBoost technique is trained using a weighted ensemble of k-NN models. For the current request, CatBoost predicts class labels that are more accurate.

Algorithm 3: Random Forest with AdaBoost
⟢ Input: – Past parking data (D) – Obtained attributes: – Occupancy rates (o_i) – Availability patterns (a_i) – Distance between parking spaces and destinations (d_i) – Cost of parking spots (c_i) ⟢ Output: – Final machine learning model for real-time parking spot availability prediction
Step 1. Subdivide D into training and testing samples
Step 2. Apply the RF mechanism on the training samples mploying mined attributs (o_i, a_i, d_i, c_i)
Step 3. Calculate the response of ML mechanism on testing sample then measure the error rate
Step 4. Using the error rate of the Random Forest model as a weighting factor and the extracted features (o_i, a_i, d_i, c_i),
train an AdaBoost model on the training set.
Step 5. Measure the errorrate of the boosting scheme on the testing sample and validate it.
Step 6. If AdaBoost error rate is lower than Random Forest error rate, use AdaBoost model as final model, else use Random
Forest model
Step 7. Apply hybrid ML scheme to forecast parking lot availability employing obtained attributes (o_i, a_i, d_i, c_i)
Step 8. Use reinforcement learning to iteratively refine the hybrid ML model in response to customer input.
Step 9. Refine and upgrade resulting version frequently to improve its precision and efficacy continuously based on consumer
demands and comments.

Algorithm 4: Algorithm for Random Forest with CatBoost

⟢

Train the Random Forest model on the historical parking data:

–

Input: Historical parking data (X_hist), corresponding labels (y_hist)

–

Output: Trained Random Forest model (RF_model)

⟢

Train the CatBoost model on the historical parking data:

–

Input: Historical parking data (X_hist), corresponding labels (y_hist)

–

Output: Trained CatBoost model (CB_model)

⟢

Predict parking availability using both models:

–

Input: Current parking data (X_cur)

–

Output: Predicted parking availability (y_pred)

⟢

Combine predictions from both models using a weighted sum:

–

Input: Predicted parking availability from Random Forest (y_RF), predicted parking availability from CatBoost (y_CB), weights (w_RF, w_CB)

–

Output: Combined prediction (y_comb)

–

Algorithm: y_comb $=$ w_RF * y_RF $+$ w_CB * y_CB

⟢

Allocate parking space to user based on the combined prediction:

–

Input: Combined prediction (y_comb), user request (X_req)

–

Output: Allocated parking space

–

Algorithm:

For each parking space p in X_req:

⋅

Calculate distance between parking space p and user destination (dist_p)

⋅

Calculate cost of parking space p (cost_p)

⋅

Calculate score of parking space p (score_p) using the formula:

⋅

score_p $=$ y_comb * w_dist * dist_p $+$ y_comb * w_cost * cost_p

Allocate the parking space with the highest score to the user.

This method integrates Random Forest and AdaBoost [22, 25], with Random Forest used to generate a diverse set of base models and AdaBoost used to enhance underperforming predictions by increasing the weights of them in accordance with the rate of error resulting from the Random Forest framework. The ultimate model can improve its accuracy and efficiency by adapting to new parking conditions with the help of reinforcement learning.

This implementation uses a hybrid algorithm that predicts parking availability in real time by combining the predictions of Random Forest and CatBoost. To predict parking availability, the Random Forest model uses past data, whereas the CatBoost model incorporates forecasts from the Random Forest model [24, 25]. Parking availability can be predicted using a weighted total of the results from both models. How well each model (w_RF and w_CB) performs on historical parking records determines their relative importance.

After a combined forecast has been made, the highest-scoring of all available parking spots is assigned to the customer [36, 39]. Each parking lot is graded based on price and proximity to the customer’s preferred parking spot. Weights for distance and expense are provided via the w_dist and w_cost parameters, respectively. The user is assigned the parking spot with the highest rating. This model combines the best features of the Random Forest and CatBoost algorithms to predict parking space availability and assign spots based on distance and cost in order to improve customer happiness [20, 24].

3.4 Numbering sequence

Sequence numbers can facilitate the process of locating a parked vehicle for drivers. A driver might be assigned a sequential number for their designated parking space. The driver can obtain this sequence number through a mobile application or a display at the parking facility [29, 31]. The sequence number facilitates the driver in locating their vehicle upon their return to the parking facility. This can optimize efficiency and alleviate frustration, particularly in expansive parking lots where locating the vehicle might be challenging [38, 42].

Parking management systems can leverage the sequence number to track spot utilization and optimize allocation. The parking management system can assign priority to each site based on its popularity by monitoring the usage of parking spaces. This information can aid drivers in locating parking spaces while also helping parking operators assign them [37]. Implementing a sequential numbering system for drivers can enhance the user experience of parking facilities, reduce the time and energy required to locate parked vehicles, and optimize the allocation of parking spaces.

3.5 Hierarchical framework

The Intelligent Parking System (IPS) is a centralized system that helps drivers discover and reserve parking spots. The Intelligent Parking System (IPS) employs a set of algorithms to determine where, when, and for how long the user needs to park their vehicle. Traditional parking [5] methods, on the other hand, just let vehicles park anywhere they like. Customers may rest assured that they will be directed to the most convenient parking spot in the lot thanks to this method. The IPS simplifies and improves parking for all users by calculating optimal parking times based on these constraints. This strategy is more efficient than picking a parking location at random [3, 16].

3.6 Centralized data

The proposed method employs a server-based intelligent parking system to enhance the distribution of parking spots and improve the user experience [8]. The concept consists of two essential components: training and testing. The technology employs a hybrid machine learning model that utilizes parking data and continually enhances its accuracy through updates. The system is trained to examine extracted factors, such as occupancy rates and availability patterns, in order to forecast real-time parking spot availability. By assimilating knowledge from previous errors and adjusting to novel parking situations, the model improves its precision and effectiveness [32]. Testing assesses the anticipated precision of the trained model when applied to novel data. The model’s viability and efficiency are confirmed by the utilization of real-time parking data. This strategy is employed by the server-based system to consistently notify users about the availability of parking spots.

3.7 Evaluation measures

a. Confusion matrix

Confusion matrices (CM) [24] are another tool for evaluating the efficacy of a classification system. It provides a more in-depth look into classification results than accuracy by displaying true positives (T $+$ ), true negatives (T $-$ ), false positives (F $+$ ), and false negatives (F $-$ ) for each class. Table 2 has the CM.

Table 2
CM to assess suggested hybrid scheme

Real class	Likelihood class		Inference
	Actually positive	Actually negative
Predicted positive	T $+$	F $-$	(T $+$ ) $+$ (F $-$ ): Positive predictions in the real class
Predicted negative	F $+$	T $-$	(F $+$ ) $+$ (T $-$ ): Negative predictions in the real class

The effectiveness of the IPS in carrying out the designated tasks is measured using TP and TN. When a driver makes a reservation for a parking spot and shows up, the prediction is a true positive (T $+$ ). When a motorist fails to show up for a reserved parking spot, it is a true negative (T $-$ ). When the model correctly predicts a reservation but the driver does not make one, we have a false positive (F $+$ ) outcome, and when the model incorrectly predicts a reservation but the driver does make one, we have a false negative (F $-$ ) event. With these four results, we can evaluate the precision, recall, and efficiency of the intelligent parking system. In order to lessen traffic and boost service quality, the proposed system provides precise estimations of reserved parking spots.

b. Accuracy

Comparisons of classification systems like hybrid-boosted ML often focus on accuracy [25]. Measures the proportion of test sample occurrences categorized successfully.

$\displaystyle\textit{Accuracy}=\frac{({TP+TN})}{({TP+TN+FP+FN})}$ (1)

c. Selectivity

In this context, “selectivity” refers to an algorithm’s ability to reliably identify positive predictions in a set of records while simultaneously reducing the number of false positives. It’s the skill of sifting through information and eliminating unnecessary details.

$\displaystyle\textit{Selectivity}=\frac{\textit{True Positives}}{({\textit{% True Positives}+\textit{False Positives}})}$ (2)

d. Sensitivity

Classifiers’ ability to reliably recognize positive cases is measured by a parameter called sensitivity, sometimes known as TP rate. When TP is compared to both TP and FN together, the ratio looks like this:

$\displaystyle\textit{Sensitivity}=\frac{({TP+FN})}{({TP})}$ (3)

e. F1_score

The F1-score is a measure of a classifier’s accuracy in machine learning that combines its precision (p) and recall (r) into a single number. It fluctuates from [0, 1], where 1 is the prime F1-score, and is the harmonic mean of ‘p’ and ‘r’.

$\displaystyle\textit{F1 score}=2\ast\frac{({\textit{precision}\ast\textit{% recall}})}{({\textit{precision}+\textit{recall}})}$ (4) $\displaystyle\text{Where }\textit{precision}=\frac{\textit{True Positives}}{({% \textit{True Positives}+\textit{False Positives}})}$ (5) $\displaystyle\text{and }\textit{recall}=\frac{\textit{True Positives}}{({% \textit{True Positives}+\textit{False Negatives}})}$ (6)

f. False Prediction Rate

The False Prediction Rate, also known as the False Positive Rate, is a statistical measure in machine learning that quantifies the proportion of negative situations that are incorrectly categorized as positive. To compute it, include both true negatives and false positives [22, 23].

$\displaystyle\textit{False Prediction Rate}=\frac{\textit{False Positives}}{({% \textit{True Negatives}+\textit{False Positives}})}$ (7)

g. False rejection rate

The False Rejection Rate (FRR) is the percentage of legitimate requests that were incorrectly rejected by a program or algorithm. FRR calculates the frequency with which a system denies a true positive [22, 23].

$\displaystyle\textit{FRR}=\frac{FN}{({FN+TP})}$ (8)

4. Result and discussion

The parking management system can benefit from hybrid-enhanced ML in numerous ways. Time and energy spent on parking an automobile are primary targets for reduction. Customers are happier, traffic moves faster, and more people can park where they want [22, 25]. The reservation system also streamlines and quickens the process of reserving parking spots, making for a better overall experience for the client.

Users’ perceptions of fairness and equality are enhanced when parking spots are distributed based on their votes cast online. Parking management efficiency can be improved by using real-time data and state-of-the-art technology to fill parking lots to capacity. These modifications have the potential to make public transit more user-friendly, sustainable, and productive.

4.1 Dataset

The dataset from UCI Machine Learning repository and Kaggle was used in this study to predict future parking potential. The data was collected from various sensors and cameras, and it includes different parking attributes such as the number of parking spaces, the occupancy proportion, and the location of the parking spaces. The datasets were analyzed using machine learning algorithms to identify patterns and trends in the data.

Dataset-1 details the parking status of a particular IPS, identified by its SystemCodeNumber (which in this case is “BHMBCCMKT01”). The four key characteristics of the information being collected are the following: “Capacity,” which represents the overall number of parking spaces in the mechanism; “Occupancy,” which represents the number of parking spaces that are currently occupied; “LastUpdated,” which represents the time stamp of the most recent update to the occupancy; and “SystemCodeNumber,” which represents the unique identifier of the parking system. This data could be utilized as part of a larger effort to analyze parking patterns or predict parking needs.

UCI machine learning and safe graph dataset-2 appears to incorporate customer parking spot distribution. Parking reservation holders have unique Ids in each column. Slot_Number is the user’s parking slot number. Second, “distance” is the user-to-parking space kilometers. Third, the parking slot request timestamp. Fourth, traffic: The volume of requests being handled when the user requested (low, medium, high). Climate: The weather when the user asks information, such as sunny, cloudy, or rainy. The day’s holiday status is stated sixth. The user’s route to the parking location is identified by Route_Id. Slot_Allocated indicates if the user received the slot.

Using this information, customers can estimate the impacts of traffic, weather, and distance on parking spot allocation and anticipate its evolution in the future.

4.2 Experimentation outcomes on dataset 1 and 2

Parking lot capacity, occupancy, and the most recent update were all included in a dataset from UCI machine learning. K-NN is a method of classification that uses a query’s nearest neighbors from the training set to make predictions about where the query should be queued. In a parking management system, K-NN could predict the likelihood that a parking place would be occupied based on the occupancy history of surrounding neighbors. By combining the results of multiple DTs, Random Forest (RF) ensemble learning produces a reliable classifier. To generate a DT, we first pick an attribute subset and training data at random. The forecast is compiled once a decision tree prediction has been made. Random forest can be used for both regression and classification. Based on the lot’s capacity, historical occupancy rates, and the time of day, Random Forest is used by parking management to anticipate whether a parking slot will be used. These algorithms’ accuracy in predicting new packing assignments is a function of the quality and relevance of the model’s attributes. Having a large amount of high-quality training data is essential for developing a model. Table 3 displays the results of ML tests on the boosted and unboosted parking lot forecast estimates from dataset 1.

Table 3
Comparison of boosting and other ML classifiers’ predictive abilities on dataset 1

Classifier	Without boosting	Added boosting mechainsm
		Adaboost	Catboost
k-nearest neighbor	0.70	0.86	0.78
Random forest	0.82	0.98	0.89
DT	0.66	0.80	0.72

AdaBoost (adaptive boosting) is an example of an aggregate learning algorithm used to handle classification problems. It uses a number of mediocre classifiers to create one reliable one. Weak classifiers trained on small samples of data are used to make the final prediction [22]. After data is partitioned into training and test sets, the code below initializes the AdaBoost classifier with 100 estimators and 20 random states. AdaBoost Classifier is fine-tuned to the training data by calling Fit (). Decision trees are used by default as the weak classifier in AdaBoost, and sample weights are adjusted so that incorrectly labeled data is given more weight in subsequent rounds. The enhanced decision boundaries achieved via boosting are depicted in Fig. 4. By summing the results of many weak classifiers, the process is repeated until the desired number of estimators is reached (here, 100). After training, test data is predicted by predict (), and the model is scored by score (). The accuracy of the AdaBoost classifier is presented for the last time.

AdaBoost instructs novice students to pay more attention to misclassified cases and less to correctly labeled ones. The results of the poor students’ predictions are then averaged and weighted according to their performance. This is done until the target number of low-performing students is combined or until the required level of training accuracy is reached. Using parking-system data, this code evaluates the relative merits of the AdaBoost and Random Forest classifiers. To improve model accuracy and learn from challenging situations, AdaBoost gives greater weight to misclassified events during training. As can be seen in Table 4, AdaBoost improved accuracy by creating a more robust model that generalized to new data.

Table 4

Effectiveness of boosted and other classifiers’ predictions on dataset 2

Classifier	Without boosting	Added boosting mechainsm
		Adaboost	Catboost
k-nearest neighbor	0.68	0.81	0.73
Random forest	0.82	0.98	0.89
DT	0.62	0.79	0.71

Accuracy is greatly improved when using a combination of algorithms, such as Adaboost and random forest or Catboost and random forest. We evaluate these methods further by utilizing formulas 1–7 from the assessment metrics section of Table 5.

Table 5

Hybrid-boosted machine learning algorithm evaluation metrics

Evaluation metric	Proposed hybrid scheme
	Adaboost with random forest	Catboost with random forest
Selectivity	0.971	0.934
Sensitivity	0.962	0.935
F1_score	0.956	0.940
False prediction rate	1.114	1.377
False rejection rate	1.103	1.387

Figure 4.

k-nn classifier using adaboosting method for dataset 2.

The measures used to rate Adaboost with Random Forest and Catboost with Random Forest can be seen in Table 5. Three metrics for measuring how well a model can distinguish between positive and negative events are selectivity, sensitivity, and the F1-score. There is a correlation between the false rejection rate and the erroneous prediction rate, which measures the proportion of falsely predicted negative cases. Adaboost and Random Forest were found to be more effective than Catboost and Random Forest.

In order to reliably predict the possibility of parking spaces, various approaches have been developed and analyzed. Table 6 compares the selectivity, sensitivity, F1-score, false prediction rate, and false rejection rate of a number of machine and deep learning methods from the literature to those of the proposed hybrid boosted ML methodology.

Table 6

Comparison of the suggested method’s performance to that of other machine learning and hybrid methods

Reference	Algorithm	Selectivity (%)	Sensitivity (%)	F1 score	False prediction rate (%)	False rejection rate (%)
Kim et al. (2018) [26]	Decision tree	91.8	90.2%	0.90	8.2	9.8
Lee and Yoon (2017) [28]	k-nearest neighbor	94.0	91.0	0.92	6.0	9.0
Al-Dmour et al. (2019) [1]	Logistic regression	95.2	93.5	0.92	4.8	6.5
Al-Dmour et al. (2019) [1]	Naive bayes	91.7	91.1	0.91	8.3	8.9
Khomh et al. (2018) [27]	Random forest	95.0	93.0	0.93	5.0	7.0
Yildirim et al. (2017) [43]	ANN	94	92	0.93	6.0	8.0
Ramanathan and Hema	Support Vector	93.0	88.0	0.89	11.0	7.0
Priya (2016) [35]	Machine (SVM)

The proposed RF-AdaBoost method outperforms the traditional scheme on the majority of assessment criteria and achieves a high prediction rate, as shown by a comparison of the response tables in Tables 5 and 6.

5. Conclusion

In order to accommodate customers’ varying needs, the suggested intelligent parking management system utilizes algorithms and data to dynamically allocate parking spots. The prediction accuracy for slot allocation was improved to 98% by combining Random Forest with AdaBoost. Such accuracy guarantees the best possible parking options for all users, all the time. Users are guaranteed their reserved parking spots thanks to the system’s verification measures, and the fee is just right taking into account the length of their stay. The complexity of the system is reduced thanks to RF, AdaBoost, and CatBoost algorithms, which increases user and operator satisfaction with the system’s optimal parking distribution. Examples are RF’s O (log n) complexity reduction and AdaBoost’s speedup and improvement in precision during processing. The intelligent parking management system optimizes parking allocation and pricing with complex algorithms, resulting in dependable and efficient parking for customers and operators.

Ethical approval

Not Applicable.

Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

Not Applicable.

Availability of data and materials

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

References

Al-Dmour

Alqudah

Al-Ani

. An intelligent parking system using machine learning algorithms. Computers & Electrical Engineering, 2019; 76: 123-134.

Azadeh

Shirzadi

Gholami-Zanjani

. Intelligent parking system: A hybrid GA-SVM approach. Expert Systems with Applications, 2019; 126: 57-69.

Bura

Lin

Kumar

Malekar

Nagaraj

Liu

. An edge based smart parking solution using camera networks and deep learning. In 2018 IEEE International Conference on Cognitive Computing (ICCC). IEEE. 2018. pp. 17-24.

Camero

Toutouh

Stolfi

Alba

. Evolutionary deep learning for car park occupancy prediction in smart cities. In Learning and Intelligent Optimization: 12th International Conference, LION 12, Kalamata, Greece, June 10–15, 2018, Revised Selected Papers 12; 386-401. Springer International Publishing.

Chavhan

Gupta

Chandana

Chidambaram

Khanna

Rodrigues

. A novel emergent intelligence technique for public transport vehicle allocation problem in a dynamic transportation system. IEEE Transactions on Intelligent Transportation Systems. 2020 Aug 11; 22(8): 5389-402.

Chawathe

. Using historical data to predict parking occupancy. In 2019 IEEE 10th Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON). IEEE. 2019. pp. 534-540.

Chong

Lim

Ibrahim

. A hybrid CNN-SVM model for intelligent parking system. Applied Sciences. 2021; 11(2): 812.

Diaz Ogás

Fabregat

Aciar

. Survey of smart parking systems. Applied Sciences. 2020; 10(11): 3872.

Skondras

Michalas

Vergados

Michailidis

Miridakis

Vergados

. Network slicing on 5G vehicular cloud computing systems. Electronics. 2021 Jun 19; 10(12): 1474.

10.

Skondras

Michalas

Vergados

. Mobility management on 5g vehicular cloud computing systems. Vehicular Communications. 2019 Apr 1; 16: 15-44.

11.

Skondras

Michalas

Vergados

Michailidis

Miridakis

. A network slicing algorithm for 5g vehicular networks. In 2021 12th International Conference on Information, Intelligence, Systems & Applications (IISA). IEEE. 2021 Jul 12. pp. 1-7.

12.

Kosmopoulos

Skondras

Michalas

Vergados

. An efficient mobility management scheme for 5G network architectures. In 2020 5th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM). IEEE. 2020 Sep 25. pp. 1-6.

13.

Fontaras

Zacharof

Ciuffo

. Fuel consumption and CO2 emissions from passenger cars in Europe-Laboratory versus real-world emissions. Progress in Energy and Combustion Science. 2017 May 1; 60: 97-131.

14.

Fukatani

Matsuzawa

Kamei

Agetsuma

Nakamura

. A method for eliminating metadata cache deallocation latency in enterprise file servers. IEEE Transactions on Magnetics. 2013 May 30; 49(6): 2504-9.

15.

Fulman

Benenson

. Establishing heterogeneous parking prices for uniform parking availability for autonomous and human-driven vehicles. IEEE Intelligent Transportation Systems Magazine. 2018 Nov 14; 11(1): 15-28.

16.

Geng

Cassandras

. A new “smart parking” system infrastructure and implementation. Procedia-Social and Behavioral Sciences. 2012 Oct 4; 54: 1278-87.

17.

Hassija

Saxena

Chamola

. A parking slot allocation framework based on virtual voting and adaptive pricing algorithm. IEEE Transactions on Vehicular Technology. 2020 Mar 9; 69(6): 5945-57.

18.

Hassoune

Dachry

Moutaouakkil

Medromi

. Smart parking systems: A survey. In 2016 11th International Conference on Intelligent Systems: Theories and Applications (SITA). IEEE. 2016 Oct 19. pp. 1-6.

19.

Huang

. Vacant parking space detection based on a multilayer inference framework. IEEE Transactions on Circuits and Systems for Video Technology. 2016 May 6; 27(9): 2041-54.

20.

Jamshed

Ali

Abbasi

Imran

Ur-Rehman

. Challenges, applications, and future of wireless sensors in Internet of Things: A review. IEEE Sensors Journal. 2022 Jan 31; 22(6): 5482-94.

21.

Kamble

Kounte

. Machine learning approach on traffic congestion monitoring system in internet of vehicles. Procedia Computer Science. 2020 Jan 1; 171: 2235-41.

22.

Karthikeyan

Kumar

Karthikeyan

. Automatic Recognition of Speaker Labels Using CNN-SVM Scheme. In International Conference On Innovative Computing And Communication. Singapore: Springer Nature Singapore. 2023 Feb 17. pp. 513-529.

23.

Karthikeyan

Suja Priyadharsini

. Hybrid machine learning classification scheme for speaker identification. Journal of Forensic Sciences. 2022 May 1; 46(3): 1033-48.

24.

Karthikeyan

Suja Priyadharsini

. A stacked convolutional neural network framework with multi-scale attention mechanism for text-independent voiceprint recognition. Pattern Analysis and Applications. 2024 Jun; 27(2): 1-5.

25.

Karthikeyan

. Modified layer deep convolution neural network for text-independent speaker recognition. Journal of Experimental & Theoretical Artificial Intelligence. 2024 Feb 17; 36(2): 273-85.

26.

Kim

Kwon

Lee

Park

. A parking occupancy prediction algorithm using a decision tree and random forest. IEEE Access. 2018; 6: 55024-55032.

27.

Khomh

Peters

Gueheneuc

. Parking slot prediction using machine learning algorithms. Journal of Ambient Intelligence and Humanized Computing. 2018; 9(2): 243-254.

28.

Lee

Yoon

. A parking-slot prediction system based on a k-NN algorithm using real-time camera images. Sensors. 2017; 17(12): 2829.

29.

Lee

Jeoung

Park

Kim

Lee

Kim

. A real-time intelligent energy management strategy for hybrid electric vehicles using reinforcement learning. IEEE Access. 2021 May 13; 9: 72759-68.

30.

Levy

Buonocore

Von Stackelberg

. Evaluation of the public health impacts of traffic congestion: A health risk assessment. Environmental Health. 2010 Dec; 9: 1-2.

31.

Lin

Chen

Chang

Chen

. SPA: Smart parking algorithm based on driver behavior and parking traffic predictions. IEEE Access. 2019 Mar 14; 7: 34275-88.

32.

Lin

Rivano

Le Mouël

. A survey of smart parking solutions. IEEE Transactions on Intelligent Transportation Systems. 2017 Apr 12; 18(12): 3229-53.

33.

Nakazato

Fujimaki

Namerikawa

. Parking lot allocation using rematching and dynamic parking fee design. IEEE Transactions on Control of Network Systems. 2022 Apr 5; 9(4): 1692-703.

34.

Nene

Mundle

Mahajan

Yeginwar

Panchal

. A study of vehicular parking systems. In ICICCT 2019–System Reliability, Quality Control, Safety, Maintenance and Management: Applications to Electrical, Electronics and Computer Science and Engineering. Springer Singapore. 2020. pp. 207-215.

35.

Ramanathan

Hema Priya

. Intelligent parking system using support vector machines. International Journal of Computer Science and Mobile Computing. 2016; 5(10): 160-167.

36.

Sahu

Naik

. A comparative study of machine learning algorithms for parking slot prediction. International Journal of Engineering & Technology. 2019; 8(1.9): 121-126.

37.

Shakya

Lee

Kim

. An intelligent parking management system based on a hybrid FCM-ANFIS model. Sensors. 2017; 17(5): 984.

38.

Stolfi

Alba

Yao

. Predicting car park occupancy rates in smart cities. In Smart Cities: Second International Conference, Smart-CT 2017, Málaga, Spain, June 14–16, 2017, Proceedings 2. Springer International Publishing. 2017. pp. 107-117.

39.

Huang

. Analysis and prediction of differential parking behaviors. In 2019 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech). IEEE. 2019 Aug 5. pp. 44-49.

40.

Ullah

Al-Turjman

Mostarda

Gagliardi

. Applications of artificial intelligence and machine learning in smart cities. Computer Communications. 2020 Mar 15; 154: 313-23.

41.

Velayuthapandian

Subramoniam

. A focus module-based lightweight end-to-end CNN framework for voiceprint recognition. Signal, Image and Video Processing. 2023 Sep; 17(6): 2817-25.

42.

Wang

Yan

Chen

. Short-term prediction of available parking space based on machine learning approaches. IEEE Access. 2020 Sep 21; 8: 174530-41.

43.

Yildirim

Sengur

Demir

. Parking lot slot prediction by using artificial neural network. International Journal of Computational Intelligence Systems. 2017; 10(1): 324-333.

44.

Zhang

Batterman

. Air pollution and health risks due to vehicle traffic. Science of the Total Environment. 2013 Apr 15; 450: 307-16.

45.

Zhang

Luo

Feng

. A comparative study of machine learning algorithms for parking slot prediction. Journal of Ambient Intelligence and Humanized Computing. 2019; 10(8): 2993-3003.

46.

Zhong

Cao

Wang

. Traffic congestion, ambient air pollution, and health: Evidence from driving restrictions in Beijing. Journal of the Association of Environmental and Resource Economists. 2017 Sep 1; 4(3): 821-56.

47.

Zhu

Liu

. Intelligent parking system using a hybrid decision tree-artificial neural network approach. Journal of Intelligent Transportation Systems. 2020; 24(3): 202-213.

An intelligent parking allocation framework for digital society 5.0

Abstract

Keywords

1. Introduction

Table 1 New developments in the field of intelligent parking systems

3.1 Feature extraction

3.2 Suggested IPS methodology

⟢ Inputs: Distance (D), Present request P, Transitional Forecast Q Outputs: Parking Label F

3.5 Hierarchical framework

3.6 Centralized data

3.7 Evaluation measures

a. Confusion matrix

Table 2 CM to assess suggested hybrid scheme

b. Accuracy

c. Selectivity

d. Sensitivity

e. F1_score

f. False Prediction Rate

g. False rejection rate

4.1 Dataset

4.2 Experimentation outcomes on dataset 1 and 2

Table 3 Comparison of boosting and other ML classifiers’ predictive abilities on dataset 1

Ethical approval

Competing interests

Funding

Availability of data and materials

References

Table 1
New developments in the field of intelligent parking systems

⟢
Inputs: Distance (D), Present request P, Transitional Forecast Q
Outputs: Parking Label F

Table 2
CM to assess suggested hybrid scheme

Table 3
Comparison of boosting and other ML classifiers’ predictive abilities on dataset 1