Enhancing Smart Parking Management through Machine Learning and AI Integration in IoT Environments

1. Introduction

The Internet of Things (IoT) has become a groundbreaking technology with extensive applications, profoundly influencing urban infrastructure and services. First introduced by Kevin Ashton in 1999, IoT refers to a network of physical objects embedded with sensors, software, and other technologies, enabling them to connect and exchange data with other devices and systems via the internet [1]. According to Gartner, IoT is one of the top 10 most important strategic technologies globally [2], with Cisco predicting about 50 billion devices connected to the internet or around 3.6 network devices per person by 2023 [3]. Making a comprehensive review [Khanna] and exploring various aspects of applications of IoT [4].

In the realm of urban management, smart parking systems exemplify the practical application of IoT. These systems are designed to address critical issues such as traffic congestion and inefficient use of parking spaces. By leveraging IoT, smart parking systems can monitor and manage parking availability in real-time, enhancing the user experience and contributing to environmental sustainability by reducing the time and fuel wasted searching for parking spots [5, 6]. The integration of IoT with machine learning (ML) and artificial intelligence (AI) further amplifies the capabilities of smart parking systems, enabling them to adapt to changing conditions and user behaviors, offer personalized services, and improve overall operational efficiency [7, 8].

The convergence of IoT, ML, and AI in smart parking systems represents a significant advancement in urban mobility and sustainable development. By collecting and analyzing data from the different sources, it can predict parking occupancy, optimize parking space usage, and implement dynamic pricing strategies [9]. This level of intelligence and adaptability is made possible through sophisticated algorithms and data processing techniques, which are essential for the effective functioning of smart parking solutions.

The need for efficient parking management solutions is underscored by the growing urban population and the corresponding increase in vehicle ownership. Conventional parking systems are often inadequate in meeting the demands of modern urban environments, leading to issues such as congestion, pollution, and frustration among drivers. Smart parking systems, have ability to provide real-time information and optimize parking resources and to offer a viable solution to these challenges [10, 11].

Previous research has demonstrated the effectiveness of smart parking systems in various contexts. For instance, the application of big data technologies in traffic monitoring and management has shown significant improvements in handling large volumes of traffic data [12]. It underscores the importance of integrating GIS and big data technologies for effective traffic data analysis and visualization [13, 14]. Furthermore, demonstrates how IoT, cloud technology, and deep learning models can predict parking space availability, reducing the traffic congestion and fuel consumption [15]. The recent studies on smart parking implementations in different cities worldwide provide a broader perspective on the advancements and effectiveness of smart parking systems in urban environments.

Barcelona has integrated various smart city initiatives, including smart parking systems that utilize IoT-enabled sensors and mobile applications [16]. These systems provide real-time parking availability, reducing traffic congestion and enhancing urban mobility. The city’s Urban Mobility Plan aims to have over 80% of journeys made via sustainable modes by 2024 [17, 18].

The SFpark project uses dynamic pricing based on real-time demand data. This approach optimizes parking space usage, reduces cruising for parking, and lowers greenhouse gas emissions. The integration of autonomous parking solutions with IoT technologies has further enhanced the efficiency and effectiveness of the city’s parking management [19, 20, 21].

Amsterdam’s smart parking systems employ a combination of sensors and cameras to monitor parking space usage. Real-time data is provided to users via mobile apps, allowing for automated payment and reservation of parking spaces. These systems have significantly improved the city’s ability to manage high traffic densities and optimize parking resources [13, 22, 23].

Singapore has developed a unified platform that integrates various parking systems across the city. This platform collects data from multiple sources, including IoT sensors and cameras, to provide real-time parking information. The system supports electronic payments and dynamic pricing to manage parking demand effectively [13, 24, 25, 26].

London utilizes Automatic Number Plate Recognition (ANPR) technology to monitor and manage parking spaces. This technology provides real-time data on parking availability and aids in enforcing parking regulations. The city’s smart parking system also integrates mobile payment options, streamlining the parking process for users and improving overall efficiency [27].

This innovative approach combines IoT and deep learning to provide accurate and reliable predictions for parking availability, showcasing the potential of advanced technologies in improving urban mobility solutions.

Defiantly intelligent vehicles and mobile robot guidance and control systems have provided valuable insights into the application of ML and AI in transportation [28, 29, 30]. The trajectory following and obstacle avoidance have highlighted the importance of precise data collection and analysis in optimizing processes and enhancing the performance of intelligent systems [31, 32, 33]. This technological paradigm shift the foundation for the development and implementation of smart systems across various domains, such as smart home [34] agriculture, healthcare, industries, and smart cities [35]. Challenges of increasing the security of IoT will be always a topic [4, 36].

This chapter explores the integration of IoT, ML, and AI in smart parking systems, aiming to address critical urban issues like traffic congestion and inefficient parking space utilization. The study hypothesizes that the convergence of these technologies will significantly enhance parking management by providing real-time data analysis and predictions.

In Skopje, Republic of North Macedonia, the implementation of smart parking systems is particularly relevant due to the city’s evolving urban landscape and increasing vehicular traffic. By adopting advanced technologies such as electromagnetic sensors, ANPR cameras, and mobile applications, can enhance its parking infrastructure and improve the overall quality of urban life [37, 38, 39]. The real case serves as a practical example of how IoT, ML, and AI can be integrated to create efficient, user-friendly, and sustainable urban solutions.

2. Design of smart parking systems

An intelligent parking system has been developed to address the challenges associated with both street parking and parking garages. Understanding the various elements that contribute to the development of such a system is crucial for comprehending how smart parking operates.

Sensors: Installed on roads and grounds, these devices help parking lot managers track customer habits and determine the suitability of available spots for different needs. They detect vehicles and update the status of parking spaces.

Cameras: Positioned in strategic locations within the parking lot, cameras gage the size and motion of vehicles. They capture images and detect motion, providing visual data to the central server.

Parking meters: These act as intermediaries between owners and customers, providing authorization and payment details to operators. They authorize parking and process payments.

Centralized server: The server receives data from sensors and images from cameras. It processes this information and manages data interactions, relaying relevant details to users via mobile apps.

Parking management software: This software communicates with stakeholders, providing real-time information on parking spaces. It updates parking information and notifies users, ensuring efficient management of parking resources.

Smart mobile apps: Mobile apps facilitate transaction handling, enabling users to find parking spaces on the street and in garages.

The relationships and interactions among these components are depicted in Figure 1, which provides a class diagram of the Smart Parking System. This diagram is crucial as it illustrates how the different elements work together to create an efficient and intelligent parking management system, highlighting the structural design and the interactions between sensors, cameras, parking meters, the central server, and parking management software.

In the context of urban environments, smart parking systems exemplify the practical application of IoT, aiming to address critical issues such as traffic congestion and inefficient use of parking spaces. These systems utilize a variety of sensors and communication technologies, such as Bluetooth, Radio Frequency Identification (RFID), and Wi-Fi, to monitor and manage parking availability in real-time [40, 41, 42]. The integration of IoT in parking management not only enhances the user experience but also contributes to environmental sustainability by reducing the time and fuel wasted searching for parking spots.

To classify smart parking systems and integrate networking and layers, we need to understand the different components and how they interact within the system. Smart parking systems can be classified into different layers: Application Layer, Network Layer, Transaction Layer, and Physical Layer. The sequence diagram in Figure 2 illustrates the architecture of integrated smart parking systems, showing how these layers interact to provide a seamless and efficient parking experience.

Application layer: User interaction with the system; Mobile and web applications for searching and reserving parking spots.

Network layer: Ensures communication between the application, parking centers, and IoT devices. Utilizes technologies like LAN, WAN, Bluetooth, Wi-Fi, 4G, and 5G.

Transaction layer: Handles transactions securely using smart contracts and consensus mechanisms. Updates the distributed ledger.

Physical layer: Consists of sensors and devices that gather and transmit data. Includes parking sensors, cameras, and other IoT devices.

Figure 2 visualizes the data flow and interaction between different layers within the smart parking system. It helps in understanding the comprehensive architecture and how various technologies and components work together to deliver a user-friendly and efficient parking management solution.

3. Data integration in adaptive learning of smart parking systems

Adaptive learning refers to systems that can adjust their operations based on data and feedback, improving over time. In the context of smart parking, adaptive learning can optimize parking space usage, predict future occupancy, and implement dynamic pricing strategies.

Smart parking systems operate by collecting data from multiple sources, to provide accurate information on parking space availability. This data is then processed using advanced machine learning (ML) and artificial intelligence (AI) techniques to predict parking occupancy, optimize parking space usage, and implement dynamic pricing strategies [43, 44, 45, 46]. By leveraging these technologies, smart parking systems can adapt to changing conditions and user behaviors, offering personalized services and improving overall operational efficiency.

The diagram in Figure 3 illustrates the model of a smart parking system, highlighting the flow of data from sensors and mobile data collectors through gateways and the cloud to the end-users. This model shows how the integration of various components and communication technologies enables real-time monitoring and management of parking spaces.

Figure 3 provides a comprehensive view of the system architecture, depicting the integration of various components and communication technologies in real-time monitoring and management of parking spaces. Here are the elements in the diagram:

• Sensors: Collect data on parking space occupancy and transmit this information to the video data collector.

• Video data collector: Gathers video data from cameras and sends it to the cloud for processing.

• Gateway: Acts as an intermediary, transmitting data between sensors, the database, and the cloud.

• Database: Stores collected data for analysis and retrieval.

• Cloud: Central hub for data processing and storage, facilitating communication between different system components.

• Access point: Provides connectivity between the cloud and user devices, allowing users to access parking information.

• User: Accesses parking information via mobile and web applications, making informed decisions about parking space usage.

• Communication technologies: Utilize Bluetooth, RFID, Wi-Fi, and cellular networks (2G/3G/4G/5G) to transmit data between devices.

• Mobile data: Enables real-time data transmission from user devices to the cloud.

By integrating ML and AI into IoT-enabled smart parking systems, we can significantly enhance urban mobility, reduce congestion, and promote sustainable development. The advanced data collection and analysis techniques allow for accurate, real-time information on parking space availability, optimizing resource utilization and implementing dynamic pricing strategies. ML algorithms analyze parking data to determine parking status and predict future occupancy, enabling smarter traffic management.

Machine learning methods in smart parking systems can be based on different types of data collected from various sources, such as images, GPS signals, environmental sensors, and time-based data. Each data type provides unique insights into parking dynamics:

SPS based on Computer vision/image processing: Utilizing camera networks, these systems extract information such as parking occupancy status, license plate recognition, and facial recognition for payment and security. They are particularly suitable for open parking lots, although they can be affected by occlusion, shadows, and lighting changes [47].

SPS based on Global positioning system (GPS): GPS guides users to available parking spots and can predict parking occupancy and traffic congestion. While suitable for open parking lots, GPS accuracy can vary, and it is less effective in indoor spaces [12]. Advantages of GPS are, highly effective for open parking lots and provides real-time location data. Challenges can be, GPS accuracy, especially in indoor or densely built environments, making it less effective for covered parking areas. This kind of example is navigation systems guiding drivers to available parking spots. Real-time traffic and parking congestion prediction [48].

SPS based on Wireless sensor networks (WSN): consists of wirelessly connected sensor nodes that monitor various environmental data facilities. These sensors are highly flexible, adaptable, and cost-effective, making them popular among SPS developers [49]. This involves connecting sensors to processing units, either through wired or wireless technologies such as ZigBee, Wi-Fi, and mobile networks (3G/4G) [50, 51, 52].

Historical and real-time reservation data: Connects end-users to SPS for tasks like data visualization, parking reservation, and payment, primarily relying on wireless connectivity like 3G/4G/5G, Wi-Fi, and Device-to-Device (D2D) communication. This method includes information such as the entry and exit times of vehicles, the total duration of vehicle occupancy, and the availability of parking spaces. Both historical and real-time data are used to improve parking management. Enables predictive analytics for better parking management and dynamic pricing strategies.

By integrating IoT with machine learning and AI, smart parking systems can significantly enhance urban mobility, reduce congestion, and promote sustainable development.

4. Methodology: Real-case

Skopje capital city of North Macedonia, has open and closed parking areas managed by ramps for access control. The public enterprise City Parking Skopje implemented 9396 parking spaces to cater to the needs of residents in regulating inactive traffic and improving parking conditions. The city’s parking areas are organized into four zones based on restriction levels and proximity to the city center: first degree (red zone), second degree (yellow zone), third degree (green zone), and fourth degree (white zone).

To analyze the parking situation in Skopje, data were collected from three parking locations within the city center: red zone, yellow zone, and green zone.

4.2 Data preprocessing and analysis

The data collected were processed to calculate the occupancy and utilization rates of the parking facilities. The formula used to calculate the occupancy at any given time is:

Z=i+Entry−ExitE1

where Z is the occupancy, i is the initial number of parked vehicles, and Entry and Exit are the counts of vehicles entering and exiting the parking area, respectively.

Тhe process of methodology and the steps of machine learning to enable smart parking is shown in Figure 4.

The process begins with collecting the parking data set. This data includes various information about parking such as entry and exit times, occupancy, and other relevant variables.

Data preprocessing: The raw data collected is preprocessed to clean and format it appropriately. This step may include handling missing values, normalizing the data, and removing any inconsistencies to ensure the data is suitable for analysis.

Data splitting: The preprocessed data is split into two sets: training data and test data. Typically, 75% of the data is used for training the model, and the remaining 25% is used for testing its performance.

Building models are made base on supervised machine learn algorithms and regression method. Given that dataset is data that is time series as the most suitable models are taken Ridge Regression, RandomForestRegressor with lagged features for time series data and, Multi-Layer Perceptron (MLP) Regressor with neural network architecture algorithms to develop predictive models.

The Ridge Regression is a linear regression which includes a regularization term in order to prevent overfitting. The regularization term, also known as the L2 penalty, adds a constraint on the coefficients of the model. The objective of ridge regression is to reduce the sum of squared residuals while also constraining the magnitude of the coefficients.

Let us denote the following:

ŷ=Xβ+bE2

ŷ is predicted target variable, X is matrix of input features, β is a vector of coeficients, and b is intercept term.

In Ridge Regression, the coefficients β are determined by minimizing the combined error between the observed and predicted target values, while also penalizing large coefficients to prevent overfitting. This is achieved by solving the optimization problem expressed in the equation:

minβ∑i=1Nyi−ŷ2+α∑i=1pβj2E3

A Random Forest Regressor is an ensemble learning method that operates by constructing multiple decision trees during training and outputting the average prediction of the individual trees. It is highly effective for regression tasks and can be adapted for time series forecasting by incorporating lagged features. It uses past values of the time series (lagged features) to predict future values. By averaging the predictions of multiple decision trees, it provides a robust and accurate forecast. This method is effective for capturing complex patterns in time series data without requiring assumptions about the underlying data distribution.

Given a time series data {yt} at time t, to forecast future values of the time series, it is used past values (lagged features) as inputs. For example, to predict y_t, the features are:

yt,yt−1,yt−2,…yt−pE4

where p is the number of lags.

For each time step t, we construct a feature vector.

Xt consisting of lagged values of the time series:

Xt=yt−1yt−2...yt−pE5

For each time step t from p to T (where T is the total number of time steps), create a feature vector Xt and the corresponding target value Yt.

For training set:

Input features:

xp+1,xp+2,…xTE6

Constructing the Training Set for each t from p to T for input features:

Xp+1=ypyp=1...y1Xp+2=yp+1yp...y2XT=yT−1yT−2...yT−pE7

Target values:

yp+1,yp+2,…yTE8

The model f predicts the future value yt̂ based on the input feature vector xt:

yt̂=fXtE9

Individual decision trees: Each decision tree hixi in the forest is trained on a bootstrapped sample of the training data.

Random subset of features: At each split in the tree, a random subset of features is selected, and the best split is chosen from this subset.

Aggregating predictions: The prediction of the Random Forest Regressor is the average of the predictions of the individual trees:

ŷ=1n∑i=1nhixiE10

where n is the number of trees.

A Multi-Layer Perceptron (MLP), consists of the input and output layer, and between them one or more hidden layers. Each layer is composed of nodes (neurons). The neurons in one layer are connected to those in the subsequent layer through weights.

Input Layer to Hidden Layer.

Input features: X=x1x2…xnT.

Weights W1

Biases b1

Transformation to the hidden layer:

Z1=W1X+b1E11

In matrix form:

Z1=W111W211⋮Wh11W121W221⋮Wh21⋯⋯⋱⋯W1n1W2n1⋮Whn1x1x2⋮xn+b11b21⋮bh1E12

Activation Function: A1=fZ1

Hidden Layer to Output Layer

Weights W2

Biases b2

Transformation to the hidden layer:

Z2=W2A1+b2E13

In matrix form:

Z2=W112W122…W1h2A11A21⋮Ah1+b2E14

Output:

yt̂=Z2E15

Evaluation metrics (RMSE, R² Score): The trained models are evaluated using metrics such as RMSE (Root Mean Square Error) and R² Score (Coefficient of Determination). These metrics help in understanding how well the model is performing in terms of accuracy and variance explanation.

The RMSE presents a measure of the differences between the values predicted by the model and the actual values. It is determined by taking the square root of the mean of the squared discrepancies between the predicted values and the actual values. The formula for RMSE is:

RMSE=1n∑i=1nyi−ŷ2E16

Where

n is the number of data points

yi is the actual value

yî is the predicted value

R² Score: is a statistical measure that indicates the proportion of the variance in the dependent variable that is predictable from the independent variables. Value of R² Score is between 0 and 1, where 1 indicates that the regression predictions perfectly fit the data. The formula for R² Score is

R2=1−∑yi−ŷ2∑yi−y¯2E17

yi¯ is the actual value

R² Score: Reflects how well the model explains the variance in the data. Higher R² values indicate better explanatory power.

These metrics aid in evaluating the accuracy and reliability of parking space availability predictions, guiding the selection of the most suitable machine learning model.

Best model by metrics results: Based on the evaluation metrics, the best-performing model is selected. This model is considered the most suitable for predicting parking availability and occupancy.

Predict parking: The selected model is then used to predict parking availability based on new data or unseen data. This step involves using the model to make predictions on parking occupancy.

Graphical comparison of real and predicted data: A graphical comparison is made between the real (actual) data and the predicted data to visually assess the performance of the model. This helps in understanding how close the predictions are to the actual values.

5. Results

The data collected provided insights into the hourly variation in parking occupancy at each location. To visualize the distribution of vehicle data across different junctions in the parking place, we utilized Python’s seaborn and matplotlib libraries to create histograms with Kernel Density Estimate (KDE) overlays. The histograms help in understanding the frequency and probability distribution of vehicle counts at various parking places, providing insights into traffic patterns and parking demand. On the x-axis, the number of vehicles is represented, while on the y-axis, the probability is shown. Figure 5 presents the visualization of vehicle distribution for different parking places: parking place 1, which is the red zone, parking place 2, which is the yellow zone, and parking place 3, which is the green zone.

These visualizations are crucial for identifying trends and patterns in vehicle distribution, which can inform decisions related to traffic management and smart parking solutions. To analyze and visualize the monthly sum of vehicle data for different parking places, we used Python’s matplotlib library to create a series of line plots. These plots provide a clear depiction of vehicle counts over time, allowing us to observe trends and variations across different parking places.

Each subplot provides a visual representation of the monthly summed vehicle counts, allowing for comparison and analysis of vehicle trends across the different parking places. The use of different colors for each parking place enhances the clarity and distinction between the plots.

These visualizations are crucial for identifying patterns in vehicle distribution, which can inform decisions related to traffic management and smart parking solutions.

Figure 6 displays the monthly dynamics of the number of vehicles in three different parking places (parking place 1, parking place 2, and parking place 3) from January 2020 to July 2021. Each subplot represents the vehicle counts for a specific parking place over time, using different colors for clear distinction.

Parking place 1 (red line): This diagram shows a steady increase in the number of vehicles over the observed period. The vehicle count starts at approximately 20,000 in January 2020 and gradually rises to about 50,000 by July 2021. The consistent upward trend suggests a growing demand for parking at this location, possibly due to increased traffic or a rise in the number of events and activities in the area.

Parking place 2 (yellow line): In contrast to parking place 1, this plot shows a more stable vehicle count with minor fluctuations until late 2020. From early 2021, there’s a noticeable increase in the number of vehicles, rising from around 6000 to approximately 18,000 by July 2021. The initial stability followed by a significant increase indicates that parking place 2 experienced a sudden surge in demand, which could be due to changes in nearby attractions, improved accessibility, or other local developments.

Parking place 3 (green line): demonstrates more variability compared to the other two parking places. The data shows fluctuations throughout the period, with peaks around mid-2020 and early 2021, and a notable rise toward July 2021. The fluctuating pattern suggests that parking place 3 may be influenced by seasonal events, temporary closures, or other periodic factors affecting its use. The increase toward the end of the observed period might indicate a shift in traffic patterns or enhanced usage.

The performance of each model was evaluated using two primary metrics: the coefficient R² Score and RMSE. These metrics provide insights into the models’ predictive accuracy and precision. Table 1 summarizes the performance metrics of the machine learning models used in the study, such as Ridge Regression, RandomForestRegressor, and Multi-Layer Perceptron, across different parking zones.

The R² score is thee coefficient of determination, indicates how well the model explains the variance in the data. The closer the R² score is to 1, the better the model’s performance. A higher R² score means the model can explain more of the variability in the target variable.

• Ridge regression: The R² scores range from 0.250087 to 0.621827, indicating that this model explains a moderate amount of variance in the data, with better performance in the red zone and lower performance in the green zone.

• RandomForestRegressor - Lag model: This model shows high R² scores, ranging from 0.725747 to 0.970134, indicating excellent performance across all zones, particularly in the red zone.

• Neural network (Multi-layer perceptron): The R² scores are also high, ranging from 0.751293 to 0.938788, showing good performance across all zones, especially in the red zone.

RMSE measures the average magnitude of the errors between predicted and actual values. A lower RMSE indicates better model performance, as it signifies that the predictions are closer to the actual values.

• Ridge regression: The RMSE values range from 5.095181 to 14.181326, indicating that the predictions have a higher average error, particularly in the red zone.

• RandomForestRegressor - Lag model: This model shows low RMSE values, ranging from 2.522930 to 5.696169, indicating more accurate predictions, especially in the yellow zone.

• Neural network (Multi-layer perceptron): The RMSE values range from 2.773242 to 5.847090, showing that the model makes relatively accurate predictions, particularly in the yellow zone.

At Figure 7 is presented the model performs. The performs on the “red” zone is with the highest R² score (0.970134) and a moderate RMSE (3.973837).

The graph shows the predicted values closely following the trend of the real values. The alignment between the predicted and real values indicates that the model effectively captures the patterns and fluctuations in vehicle numbers over time.

Occasional spikes and dips in the real values are also reflected in the predicted values, demonstrating the model’s ability to respond to variations. However, some discrepancies exist, particularly in the peaks and troughs, which is expected in any predictive model.

Figure 8 illustrates the performance of the predictive model for parking place 2 (Yellow Zone) by comparing the predicted values (dark yellow line) against the real values (light yellow line) over time. The model performs well on the “yellow” zone, with a good R² score (0.885717) and the lowest RMSE (2.522930), indicating the most accurate predictions.

Both figures show that the predicted values closely follow the real values, but Figure 7 has occasional larger discrepancies in peaks. Figure 8 demonstrates stronger alignment in capturing peaks and troughs. Both models are highly useful for real-time management, but the model for the Red Zone (Figure 7) shows slightly better overall variance explanation, while the model for the Yellow Zone (Figure 8) has a lower average prediction error, indicating more precise predictions.

Figure 9 illustrates the performance of the predictive model for parking place 3 (Green Zone) by comparing the predicted values (dark green line) against the real values (light green line) over time.

The model performs the worst on the “green” dataset, with the lowest R² score (0.725747) and the highest RMSE (5.696169). The graph shows that the predicted values (dark green line) generally follow the trend of the real values (light green line). However, there are noticeable discrepancies, particularly around the peaks and troughs, where the model fails to capture some of the extreme variations in the real values. Despite these discrepancies, the model captures the overall pattern and central tendencies in vehicle numbers over time, indicating less accurate predictions compared to the other datasets.

6. Proposed development of a smart parking solution for Skopje for future work

Based on the data collected and the observed peak parking hours, a smart parking system utilizing mobile applications is proposed. The model follows the example of Barcelona [13, 53], where wireless sensors provide real-time parking status updates through a mobile app. This system includes:

• Electromagnetic sensors at each parking spot to detect occupancy,

• ANPR cameras for vehicle identification,

• Infrared cameras at entry and exit points for monitoring,

• A mobile app for users to check availability, make reservations, and manage payments.

Implementing this smart parking system will reduce traffic congestion during peak hours, streamline the parking search process, and provide real-time parking information to drivers. The intelligent system will modernize vehicle control and reduce the need for human labor, making it an efficient and sustainable solution for Skopje’s growing urban infrastructure.

The development of the smart parking system involves installing wireless sensors and providing real-time updates on the status of parking locations through a mobile application. As the most appropriate and acceptable, I suggest that it be applied to the selected locations in Skopje.

The following Figure 10 shows the architecture of the proposed smart parking model.

The way this proposed smart parking system works is as follows: first, the user requests information from the mobile application. In this case, the user can request where the parking is available, the location of the available parking space, payment information, etc. Sensors placed in each parking space detect whether the parking space is available or unavailable and transmit the information wirelessly to the database and then to the user. The mobile application then displays information stored from the database in the server and resolves the user’s query in real time. Vehicles can be parked in the available space and the user can pay for the parking space and locate the car at any time using the mobile application.

The electromagnetic sensors are installed in each parking space, in each of the parking locations, which detect the status of each parking space and send all the information about the occupancy to the central management unit. To deliver the occupancy status of each parking space, sensors are connected to an IoT network. The sensors work on batteries and continuously register the presence of a vehicle in the parking space and wirelessly send the data to the receiver. A receiver operating in the 865–867 Mhz range updates this server occupancy data to the control center over the Internet using a wired network, as shown in Figure 11.

The identification of the vehicles arriving or leaving the parking lot is done using an ANPR camera (Figure 11). ANPR stands for Automatic Number Plate Recognition and contains the technology to identify a vehicle by number plate. With the implementation of automatic license plate recognition cameras, the technology can scan a license plate and forward the associated action to the management system. This allows for a modernized version of vehicle control and limits the need for human labor given the stand-alone nature of ANPR technology. In addition, the parking lot has cameras (based on infrared technology) with the ability to zoom, located at the entry and exit points to monitor the parking zone. Any unauthorized entry of the vehicle that does not go through plate recognition can be easily verified by the operator.

Users can check the availability of the parking zone, as well as get information about the location and availability of parking locations through the mobile application installed in their mobile phones. The mobile application also shows users the occupancy of parking zones in real time.

As described above, the mobile application offers two ways for drivers to select a desired parking location, by searching for parking locations through the nearby button and by entering the address of the desired parking location, as shown in Figure 12.

After they select the desired address, the coordinates are sent to the management system which then updates the data in real time. If the user selects the nearby button, the GPS system locates the user’s current location and the application displays the distance and availability of parking locations as shown in Figures 12 and 13.

7. Conclusions

The integration of IoT, machine learning, and artificial intelligence within smart parking systems offers a promising solution to the growing challenges of urban mobility and parking management. This chapter has demonstrated how leveraging these advanced technologies can optimize parking space utilization, reduce traffic congestion, and enhance the overall user experience.

The data analysis and machine learning models applied in this study have provided valuable insights into parking occupancy patterns and the effectiveness of predictive algorithms. The RandomForestRegressor with lagged features exhibited superior performance across all parking zones, making it the most effective model for predicting vehicle counts. The Neural Network (Multi-layer Perceptron) also showed good performance and is a viable alternative, while Ridge Regression, though useful, did not perform as well as the other two models, especially for the data of the green zone.

The deployment of such an intelligent parking system in Skopje is poised to modernize the city’s urban infrastructure, reduce the environmental impact of traffic congestion, and promote sustainable urban development. By continuously monitoring and managing parking resources, the system can adapt to dynamic conditions and user behaviors, offering personalized and efficient services to drivers.

This chapter highlights the critical role of smart technologies in addressing urban challenges and sets the stage for further advancements in smart parking systems, paving the way for smarter, and more efficient, and sustainable cities.