An efficient AI model for highway vehicle speed limit violation detection and real-time law-enforcing functionality

a. Problem statement

Surpassing the speed limit on highways is a major contributor to fatal accidents. Numerous research papers over the years have proposed approaches and tactics to mitigate this issue [24,25,26]. These techniques can substantially mitigate the issue. Nonetheless,

1. None of these provides a mechanism for pre-screening unpaid speeding tickets to inhibit their access to the highway.

2. No traffic management office has implemented a consistent procedure for issuing summonses to speed limit violators.

3. The initiatives to deter reckless driving diminish in efficacy when it requires several days or weeks for the relevant authority to summon the offender.

b. Hypothesis

A comprehensive conceptual framework can be established to tackle instances of speeding offenses by motor vehicles utilizing IoT enabled AI technology [27,28,29,30]. Additionally, it would utilize a sophisticated SA technique to improve detection precision and data integrity. This design would facilitate the development of a very precise real-time system for recognizing and notifying specific instances of vehicles exceeding the speed limit. Moreover, it would enable the prompt enforcement of instant fines, such as citations, at toll exit booths, stations, or traffic offices. This system would restrict autos with unresolved speeding infraction summonses from utilizing the route [31].

c. Research questions

The subsequent inquiries emerge from the necessity to develop a sophisticated traffic system [32,33,34] for monitoring highway speeders and enforcing immediate sanctions for habitual overspeeding:

1. How can we develop and construct a precise, real-time IoT based AI enabled system for detecting and warning vehicle over-speeding situations on highways, with minimal initial and recurring costs?

2. What is the optimal method for developing and executing an effective SA system that uses intelligent wireless scanning to accurately deter auto-mobiles with outstanding speeding summonses from entering the highway?

3. How can we integrate real-time databases to ensure that when a driver exceeds the speed limit, they are promptly summoned for accountability at the toll booth, station, or traffic regulation office upon exiting the highway?

d. Objectives

This study seeks to create a comprehensive conceptual model utilizing IoT technologies, big data analytics, and deep learning methodologies to identify instances of vehicles surpassing speed limits on highways [35,36,37]. The objective is to diminish the incidence of fatal traffic accidents and enhance awareness among vehicle owners and operators. Additional objectives can be emphasized on the need for the smart traffic overspeeding monitoring and law enforcement system:

1. IoT gadget can be deployed in vehicles to properly identify and report speed offenses.

2. A system enabling law enforcement to promptly dispatch text messages, including fines and summonses, upon detection of a traffic violation. The penalties are shown on the corresponding IoT device within the car. This system employs effective SA and a smart AI approach. This would promptly deter and caution the motorist.

3. Vehicles having three or more outstanding traffic citations can be swiftly and efficiently addressed to discourage drivers from repeatedly violating the established speed limit.

e. Novelty analyses and contributions

Currently, it is a prevalent practice to overseed highways. This results in several fatal traffic accidents. Consequently, infractions of highway speed limits must be identified immediately, and offenders should be penalized without delay [38]. Significant effort has been invested in developing an IoT device that employs various machine learning and AI methodologies to address these issues instantaneously. Among the various methodologies, a prevalent strategy is the integration of you only look once (YOLO) with recurrent neural network (RNN), long short-term memory (LSTMs), and convolutional neural networks (CNNs) for time-series tracking [39, 40]. The inefficiency and imprecision of the conclusions produced by these methodologies in real-time are attributable to certain flaws. The numerous issues with YOLO-based real-time tracking include its limited intrinsic tracking skills, difficulties in managing small or over-lapping objects, and insufficient object localization. In object detection, the YOLO model exhibits challenges with output precision despite its acclaim for temporal efficiency. The vanishing and exploding gradient issues arise concurrently in RNNs. During traversal of the network, the gradients employed in the learning phases of backpropagation may either significantly diminish (vanish) or substantially amplify (explode) [13]. Consequently, the RNN model encounters significant challenges in learning, resulting in unreliability and inconsistent outcomes. This is the reason LSTM performs effectively with sequential data. However, it also possesses other downsides, such as being computationally challenging and necessitating substantial time and financial resources for training. Moreover, overfitting is a prevalent problem, particularly with limited data. The computational inefficiency of CNNs, particularly in comparison to deep networks, constitutes a significant drawback of CNNs. Convolutional Neural Networks necessitate substantial processing resources due to the extensive number of convolutional operations and parameters involved [41,42,43]. The constraints established by the spatial hierarchy present an additional challenge for Convolutional Neural Networks. While they eliminate specific geographical data, pooling layers diminish feature map dimensions and confer translation invariance. CNNs frequently encounter difficulties with generalization and necessitate substantial quantities of labeled training data. Insufficient labeled datasets can lead to overfitting in CNNs, particularly in contexts such as uncommon disease diagnosis [14]. The issues with existing approaches can be resolved by integrating SA with artificial neural networks (ANNs), as outlined. A substantial body of research indicates that ANNs may readily adapt to novel activities and situations without necessitating reprogramming. Moreover, ANN is more straightforward to construct and implement for real-time data processing. Only one iteration is employed in the SA process. As a result, SA is efficient and can yield optimal results with significant accuracy and adaptability [44,45,46]. Consequently, the optimal approach to address the existing issues with the proposed hardware configuration is to integrate ANNs with SA. This study significantly enhances prior attempts at real-time data collection by the utilization of a highly efficient motion sensor. It facilitates the effective transmission of processed challan information and speed data to the roadside device. The primary objective is to minimize data inaccuracies during the collection, processing, and transmission phases of the method.

a. Comparison of the proposed technique with other existing traffic management systems based on their efficiencies

The performances of the proposed integrated strategies are compared with existing approaches from contemporary literature to further evaluate the effectiveness of the suggested method, as presented in Table 3. Table 3 presents a comparison that encompasses the intelligent and precise monitoring of vehicle speed on highways, alerts the driver when attempting to surpass the speed limit, imposes fines for repeated violations of the speed limit, ensures a security framework for data integrity, and offers a detailed cost estimation.

Table 3:

A summary of various challenges of cloud resource provisioning and their mitigation

Publication Year/reference		Monitoring vehicle speed smartly and accurately	Worn the driver for violating the speed limit	Imposing a penalty for continuous speed limit violation	Security and integrity of traffic data	Cost efficiency of the used technique
2017	[13, 15]	○		○
2018	[33]		○	○		○
2019	[21, 25]			○
2020	[1, 3, 9, 12, 17, 39, 45]					○
2021	[2, 26, 38]		○	○
2022	[11, 14, 37, 46, 47]		○	○
2023–24	[7, 8, 9, 28, 30, 32, 41, 49]		○	○
2025	[5, 20, 34, 50]		○	○
Our Work (2025)

Table 3 illustrates that the suggested integrated technique effectively addresses all aforementioned characteristics. The alternative methodologies in the current literature are either incapable or only partially capable of addressing all these issues in a cohesive manner. A comprehensive study of the strengths and weaknesses of the current body of work is necessary to identify the existing gaps in the literature. This study aids in determining the suitable strategy to address the current gap in the literature. This section presents a detailed analysis of the diverse existing methods for regulating traffic-related offenses on highways, based on contemporary literature, highlighting their strengths and weaknesses in the following Table 4.

b. Impact of other existing intelligent systems on driver attitude

Contemporary car speed monitoring systems gather data using speed measurement sensors, evaluate the information, and issue alerts for exceeding speed limits [5, 13, 14]. Furthermore, most of these existing solutions lack a communication route with the traffic department. The current research provides limited insights into the integration mechanism of the traffic department; nonetheless, these approaches are insufficient to deter drivers from excessive expenditures and dangerous driving habits. As a result, these strategies do not evoke any apprehension among speeding motorists. Consequently, these advanced systems exhibit negligible influence on drivers who engage in hazardous behavior and persistently attempt to exceed speed limits in areas lacking traffic surveillance cameras. However, the increased warning generation due to speeding offenses on a certain highway is advantageous for cautious drivers. It aids them in sustaining an appropriate speed within the established limit, therefore mitigating the risk of adverse situations and serious accidents. Therefore, to alleviate irresponsible and lavish driving, these tactics must be amended through cooperation with traffic authorities.

c. Research gap analysis

The literature study indicates that current approaches are inadequate for a real-time monitoring system capable of detecting instances of overspeed violations. Nevertheless, there have been occurrences in which data gathered from highway traffic are unanalyzable due to ambient noise or substandard data quality [25, 26]. Consequently, a real-time IoT-driven AI model is essential for monitoring and regulating excessive speed in highway traffic. Further research and development are necessary to guarantee both accuracy and data integrity concurrently. Importantly, none of the proposed methods or strategies provides a structure or framework to restrict cars that have consistently exceeded speed limits from entering or accessing highways or expressways [28]. The literature review reveals a lack of studies on the real-time synchronization of overspeed violation data from multiple vehicles to issue summonses to drivers who violate speed limits at exit toll booths or stations [28, 29]. Furthermore, the study neglects to consider the critical issues of implementation and maintenance costs.

a. Design of the proposed technique based on earlier information

The literature review indicates a notable inadequacy in the existing system for the real-time detection and penalization of speeding by the appropriate authorities [23,24,25,26]. In the absence of a structure to impose sanctions for speeding, drivers are continually motivated to engage in this conduct. A survey was conducted concurrently to collect and evaluate the demand from various users, together with their comments and feedback concerning the necessity of the proposed device. Based on the literature analysis and public feedback, the suggested IoT sensors have been created with a compact rechargeable lithium battery and Bluetooth connectivity. The hypothetical placements of the proposed sensors around highways and toll gates are illustrated in Figures 4A–4D.

Figure 4:

b. Design of the proposed ANN and SA

ANNs have acquired significance in numerous fields due to their functional similarities with biological systems. ANNs excel at generalizing ambiguous inputs, demonstrating fault tolerance, facilitating extensive parallelism, and solving non-linear problems. ANNs aim to offer mathematical frameworks for data processing that leverage existing knowledge bases. ANNs models can produce very precise solutions for situations that defy traditional methods and are mostly understood through empirical data. These models are created by experimental methodologies [13]. The internal design of the ANN is flexible in response to prevailing conditions due to a learning mechanism, and its ability for swift data processing and fault tolerance is ascribed to its significant parallelism. The obtained data are generalized to improve their relevance to new or varied datasets. The choice to employ ANN for the model development of this project was influenced by these factors [12, 13]. The applications of ANN in speed estimation for moving vehicles are widely employed because they successfully calculate vehicle speed while reducing errors and noise from the input data. The ANN operates best with a wide range of input data formats, including different types of sensor data. This study will utilize ANNs to develop a highway speed prediction model, which will subsequently be assessed using different personal vehicles.

At this phase of implementation, the proposed integrated methodology is amalgamated with an innovative design, anANN, and a SA model [25,26,27]. Initially, the proposed architecture of an ANN is utilized to classify sensor data obtained from the vehicle's internal IoT components. The proposed architecture of the ANN comprises 150, 300, 300, and 30 hidden layers operating collaboratively. Seventy feature vectors were selected for this implementation, considering the dimensions of the input layer. The proposed ANN thus has three classes in its output dimension. Furthermore, the loss function distinctly intersects the entropy value in this implementation. A 10% dropout rate in each layer has been investigated to mitigate the overfitting problem. The adaptive movement estimation (ADAM) optimizer function and the rectified linear unit (ReLu) activation function are employed for each neuron. This model was created using Python through the KERAS-Application Programming Interface (API) with TensorFlow operating in the background. Figure 5 illustrates that the proposed ANN model adheres to a conventional structure.

Figure 5:

The first and second iterations of the projected ADAM model are represented by the exponential gradient (E_x) and the squared gradient (S_x) The hyperparameters ∝₁ and ∝₂ ∈ (0,1) control the rates of exponential decay. Here,

The inputs in this case are (1,…x₁,…x_n), and the function g(x), which adds up all the inputs biased to $\widehat{E_x}$ and $\widehat{S_x}$ , calculates the output neurons. These are employed to further estimate the first and second iterations.

Eq. (5) represents the final update parameter as γ (x + 1), which represents the final weight parameters, and γ (x), which represents the previous weight parameters. If μ represents the learning rate, then ɛ (epsilon) is extremely small, preventing any division in this process by ‘0”. If ∂, is any constant value, then we have,

All the hyperparameters in this projected model are set at μ = 0.001, ∝₁ = 0.989, ɛ = e⁻⁰⁹ and decay = 0.

b.ii. Role of the SA for overspeed detection

SA, as an optimization technique, employs numerical models that accurately represent the data. A randomly generated model is systematically refined until it conforms to the data [22, 23]. Our methodology employs SA to identify and document overspending data derived from vehicle speed information obtained from the onboard speed sensor, alongside the threshold speed limit gathered from roadside Bluetooth sensors. The suggested SA approach concurrently assesses if any vehicle exceeds the specified speed restriction on a particular highway. It also considers the traffic conditions and congestion levels of any highway, using calculations derived from this data. The congestion data are solely collected from the roadside Bluetooth device. A speeding notice is generated and dispatched by a thorough assessment of all pertinent factors. Furthermore, after the third consecutive speed limit violation on a certain highway, a citation is generated and sent to the adjacent roadside devices through Bluetooth communication. This information is then communicated to the traffic department, requiring the driver to pay the penalty at the next toll plaza or traffic control office. This section outlines the capabilities of the proposed SA method to enhance the precision in identifying instances of overspeed. It is essentially a method that uses probability to determine the ideal value. It is a metaheuristic approach utilized to evaluate the incidence of speed limit violations. The SA method for ascertaining overspeed from the input vehicle speed data is detailed in Algorithm-1.

After calculating the penalty values for different scenarios using Algorithm-1, the final summons, or challan, is generated based on these penalty values. The resultant summon is thereafter displayed on the linked screen, accompanied by all pertinent information as illustrated in Figure 6.

Figure 6:

SA may optimize many hyperparameters in any AI model and improve its architecture. To improve the performance of the proposed ANN model, the effectiveness of several hyperparameters, including ADAM and ReLU, is optimized using the recommended SA method. We employed SA to identify and document overspending information derived from vehicle speed data obtained from the onboard speed sensor and the threshold speed restriction provided by roadside Bluetooth sensors. In the domain of neural networks, any SA technique can explore the hyperparameter space, facilitating adjustments that improve performance. The suggested SA method concurrently assesses if every vehicle on a certain highway surpasses the established speed limit. Thus, the proposed SA method aids in determining the ideal hyperparameters that may be modified to improve the accuracy of overspeed detection. It also considers the traffic conditions and congestion levels of every highway, employing algorithms derived from this data. Furthermore, it contributes to reducing the mean squared error of the proposed integrated ANN model and enables accurate model training.

Algorithm-1:

Simulated Annealing for Vehicle Overspeed Detection

Input Data: →Number of speed limit violation, epochs;

Result:→ Generating Summon Accurately after violating the speed limit contagiously 3 times;

1. Let S ←S0 be the initial solution where the driver violates the speed limit for the first time;

2. T←epochs;

3. While T> 0 do

a. Calculate speed limit value from the road sensor and set it as the maximum speed limit (m);

b. While number of violation (n) <3 do

(i) Select any other recorded speed (xj) at the random basis in the next iteration;

(ii) If xj > m

If the root is congested enough

Calculate the penalty value (Pj);

End if

If the root is not congested enough

Calculate the penalty value (Pk);

End if

Else if $\text{e}\hspace{0.17em}\frac{P_j-P_k}{k.T.}>\text{Random}(0,1)$ , where k is any random integer

Calculate the penalty value (Pf);

Penalty ← Pf.

Break;

End if

End if

T ←T-1;

End While

End While

c. Design and incorporation of the proposed error control technique

As indicated in the introduction, data errors play a crucial role in any transportation system. The data are sometimes compromised or lost during transmission via unsecured networks due to variables such as channel congestion, restricted capacity, and environmental interference [16, 18]. The suggested integrated technique involves transmitting overspeed information from any vehicle to the roadside Bluetooth device and relaying highway-related information from the roadside Bluetooth device to the vehicle. Consequently, data loss may occur, allowing the true speed limit violation to evade punishment. Consequently, in the last stage of this implementation, an error control method is developed and incorporated with the suggested approach to guarantee the accuracy of data during its transfer from the local device to the roadside storage or from the roadside storage to the traffic department. The design of the proposed methodology and its integration with the suggested method are depicted in Algorithm-2. In this instance, we have utilized the final challan information as input and transformed it into a binary string to produce the error control bits during the inclusion phase.

Algorithm-2 produces the ultimate error control string by doing multiple rounds of Exclusive OR (XOR) and concatenation operations on various sets of input strings. Once the final error control bits generation and concatenation procedures are completed, the information is saved in a local IoT device. It can then be sent to the roadside storage device using a Bluetooth network.

d. Role and importance of the display device

The resulting summons or wordings are displayed on the display device when all the data has been processed using the recommended integrated approach. The major role of the display device is to notify drivers of the vehicle's speed and alert them when they surpass the authorized speed limit, allowing for speed regulation and the prevention of accidents or unfavorable situations. This display apparatus must be integrated into the vehicle to communicate information on speed limit violations, traffic infringement penalties, or alerts from the designated internal IoT device [13, 38]. The proposed method stipulates that the internal IoT device will transmit a warning to the display device upon any violation of speed restrictions, prompting it to exhibit a message. After three consecutive violations of the designated speed restriction, the display gadget generates a citation for the motorist. Figure 6 depicts the arrangement of the proposed display device.

Figures 4 and 6 illustrate that the internal IoT device interfaces with external sensors situated on the roads via Bluetooth devices in auto-pair mode. Bluetooth devices must be installed at every 1 km interval along the roads to capture and retain vehicular information. In the occurrence of a device malfunction, the alternative device can access the backup data with few or no complications. Conversely, communication delay presents a substantial challenge in these situations [11,12,13]. The employed Bluetooth device functions on a many-to-one communication mechanism, featuring a total of eight accessible channels. This enables concurrent communication with a maximum of eight vehicles, rendering it appropriate for highway applications. The minimum data transfer rate of a Bluetooth device in a many-to-one or one-to-many setup is 800 kbps. This signifies that each channel of a many-to-one Bluetooth device can send data at a rate of 100 kbps. The proposed framework stipulates that the maximum size of each handshaking message shall be 5 KB. Consequently, transmitting a 5 KB message will need a maximum of 0.05 s. The operational range of the Bluetooth device employed in our proposed architecture is 100 m. A vehicle capable of reaching a maximum velocity of 240 km/hr may traverse 3.3 m in under 0.05 s. Even in particular circumstances, when the network speed diminishes to 4 kbps, it remains capable of communicating with a high-velocity car traveling at 240 km/hr. Consequently, in both instances, the proposed architecture may adeptly address the challenges with efficiency.

e. Retrieval of original information

The acquisition of unique information is crucial for any transportation system [46]. The suggested integrated system utilizes data from embedded speed sensors within the vehicle, which are subsequently analyzed through the combination of SA and ANNs to reliably identify specific speed limit infractions on designated highways. The output is further processed using the suggested error control technique, to which additional error control bits are incorporated for managing various sorts of data errors that occur during transmission. These error control bits also facilitate the creation of a backup system in the event of unintentional data loss. Consequently, the processed information (i.e., following the application of the proposed [SA] and [ANN]) must be extracted at the receiving end in its original format from the error-controlled string (FinErr). At this level, the utilized input string is the final error control string (FinErr). Each trio of 8-bit strings is consecutively delineated [8]. Perform the XOR operation on the initial two strings as per Algorithm-2, and evaluate the outcome against the third string. If the match is incomplete, eliminate both strings from the third string and thereafter compare the resulting strings with the original string to identify the appropriate combination of strings. Execute analogous methods for the residual segment of the string and consolidate all the extracted strings to reconstruct the original string. After extracting the original string, it must be reverted to its initial textual format for display on the device. Consequently, the original data, in the form of a challan or other documentation, is retrieved at the receiving end.

f. Key challenges in implementing the proposed model

Several challenges have arisen throughout the execution of the proposed model. The principal obstacles encountered in this research were related to the establishment of the hardware configuration. The proposed model is designed for detecting high speeds on highways. Consequently, we needed to obtain authorization from both the traffic department and the local police station to operate various car types at a designated speed during data collection. Additionally, the selection of sensors presented a significant challenge due to their varying ranges. Consequently, it was exceedingly challenging to determine which type of sensors would yield the most efficient results. The selection and deployment of Bluetooth devices presented a significant challenge in this research endeavor. The data reception and transmission capabilities of any Bluetooth device are significantly constrained. Accurate distance measuring is essential to yield precise results and reduce transmission errors. Consequently, the selection of an appropriate location and devices proved to be a formidable undertaking. The integration of communication between the speed detection sensors, display device, and the roadside Bluetooth apparatus was significantly transformative in this research as well. However, we have effectively surmounted these hurdles through our persistent efforts and the help of local authorities to create and evaluate the effectiveness of our suggested model across various dimensions.

g. Real-world deployment of the projected model

Assessing the real-time functioning of any IoT device is essential [50]. The prototype of the suggested model was installed in five different models of cars, such as Suzuki Swift, Tata Nexon, Hyundai Grand i10, Skoda Kylac, and Honda City 2023 models, and tested accordingly. All vehicles are evaluated on a traffic-free road in Panditwari, Dehradun (Uttarakhand) within a speed range of 70–100 km/hr due to specific traffic restrictions. The test runs for all these vehicles were undertaken for 2 hr daily over a period of 2 weeks. The Wireless Ultrasonic Sensor (Uno Minda 7687Q-028-11-M01 SPEED SENSOR For-BAJAJ-DISCOVER 125ST) was used to record the vehicle speed. A MacBook functions as the workstation for collecting vehicle speed data. Communication between the roadside Bluetooth device and the workstation has been established by Bluetooth signals. The used Bluetooth device is a class-1 type of device which can cover, that is receive data or transmit data within a range of 100 m. The MICROUSB Compatible with Regold CR2016 Lithium Coin Button Battery 3v has been used in IOT sensor during the data collection operation. The installation of the Bluetooth device at the roadside and the integrated module was performed similarly to the representations in Figures 4A–4D and Figure 5. The results will be further analyzed in the results section to clarify the effectiveness of the proposed technique across multiple aspects.

a. Data set preparation

During the data collection, Wireless Ultrasonic Sensor (Uno Minda 7687Q-028-11-M01 SPEED SENSOR For-BAJAJ-DISCOVER 125TH) were used to collect the speed data in JSON file format. After the completion of the test run of five different cars that is Suzuki Swift, Tata Nexon, Hyundai, Grand i10, Skoda Kylac, and Honda City 2023 models, along with the installed IOT device, the captured data stored in the local storage and preprocessed with the help of cleaning process to handle the missing values and normalized the data to create suitable input data that produces the accurate and desired output. Various libraries are added to make the processing unit very efficient and to support various input file formats. We considered various lighting conditions, including morning, evening, and afternoon, as well as sunny, overcast, and rainy weather, when preparing the dataset to obtain the input data under various circumstances. The test drive was conducted in low-density traffic conditions due to restrictions imposed by the local traffic authority.

b. Hardware setup

Wireless Ultrasonic Sensor (Uno Minda 7687Q-028-11-M01 SPEED SENSOR For-BAJAJ-DISCOVER 125TH) has been installed inside the car to collect the overspeed data. The IoT device that has been proposed is constructed using an LM339 from ON Semiconductor, which is a Low Power Quad CPU, for the purpose of processing data, an Liquid Crystal Display (LCD) Unit (2.4 inches) for displaying the information, and an 8 GB storage unit for storing the processed data. Aside from that, the power has been supplied by a lithium battery with a voltage of 5 V and a life span of 2–3 years. However, in addition to that, a Many-to-One Bluetooth device and vehicle speed sensors (VSS) are utilized in order to record the speed of the vehicle at regular intervals. Every single component of the hardware is put together and put through its paces throughout the inspection process.

c. Software setup

In the software setup part, we have used Python within the KERAS API for implementing the projected ANN and SA part. Apart from that, simple Python programing has been used to develop and incorporate the proposed error control bits generation and the incorporation process. Simple integrated networking software and Point-to-Point protocol (PPP) is used to handle the Bluetooth devices. Finally, all the programs are implemented in windows 2010 platform.

d. Scalability and cost implications

The cost calculation of the proposed architecture has been conducted based on Indian market prices to assess its affordability. It comprises two modules: the cost of the integrated module, which will be situated within the car, and the cost of the many-to-one Bluetooth device, which will be positioned roadside. The expenses associated with the implementation, installation, and maintenance of both modules are presented in Table 5.

Consequently, the cost assessments in Table 5 indicate that both components for constructing the suggested model are economically viable for road users and transport authorities to implement the predicted design in practice. The display device is compact, and the Bluetooth device is highly portable and adaptable according to the available space within the vehicle or roadside, allowing for easy real-time deployment.

a. Evaluation to attain higher accuracy with various performance matrices

The effectiveness of the proposed combination, along with other AI techniques, in accurately detecting overspeed in formation has been tested in this sub-section. Here, the proposed integrated technique is designed by incorporating a unique ANN-based approach. Therefore, the capacity of the proposed integrated technique has been calculated and compared with some well-known AI techniques such as weighting and clustering outliers and internal cases detection (WCOID) [5], case method based on clustering and outliers and internal detection (COID) [27], Condensed Nearest Neighbor Rule (CNN) [7], RNN [26], R-CNNs [1, 9, 11] and instance-based learning (IBL) [21]. The accuracy percentage [9, 30, 49] for detecting overspeed information using different AI-based algorithms has been computed using the following Eqs (6) and (7).

Where TP, TN, FP, and FN represent True Positive, True Negative, False Positive, and False Negative, respectively [5, 30]. An efficient system increases the accuracy percentage of detecting the overspeed information, or vice versa. Further plotting of the percentage of accuracy provided by the proposed combo as well as the other AI techniques, can be found in the following Figure 7.

Figure 7:

The details of Figure 7 is further elaborated in the following Table 6 to make the accuracy percentage and the efficiency of the proposed integrated technique more visible.

Any deep learning model may be assessed for efficiency using many performance metrics, including precision, recall, F1 score, and area under the ROC curve (AUC) scores [9, 30, 49]. A deep learning-based strategy is deemed efficient in accurate prediction if it demonstrates superior precision, recall, F1 score, and AUC scores, or conversely. Consequently, the precision, recall, and F1 scores are computed using Eqs (8)–(10) as referenced in [9, 30, 49].

The probability that a classifier favors a randomly chosen positive instance over a randomly chosen negative instance is denoted by the Receiver Operating Characteristic-Area Under the Curve (ROC-AUC) score. It evaluates the effectiveness of a binary classifier across all possible classification levels [49]. The performance analysis based on precision, recall, F1 score, and AUC scores for correct prediction is illustrated in Figure 8.

Figure 8:

Table 7 that follows provides more details on the aspects of Figure 8, making the various AI methodologies more visible in terms of Precision, Recall, F1-Score, and AUC-Score, as well as the efficacy of the proposed integrated strategy.

Figure 7, Table 6, Figure 8, and Table 7 demonstrate that the suggested procedure yields superior Accuracy Percentage, Precision, Recall, F1-Score, and AUC-Score compared to other existing AI methodologies. These facts justify that the suggested strategy is efficient in providing superior prediction accuracy compared to existing methods. Consequently, the proposed combination surpasses other current AI methodologies for detecting speed limit infractions, as distinctly seen in Figure 7, Table 6, Figure 8, and Table 7. These findings indicate that the proposed combination can achieve both our first and second research objectives.

b. Analysis of its capacity to protect against data loss

Data loss is a critical feature that can occur during any transmission procedure. Data can be lost, either partially or fully, due to many factors such as limitations in the transmission channel, environmental noise, constraints of the network being used, channel congestion, and other similar reasons [5, 18]. Hence, in any traffic management or monitoring system, the loss of transmitted data poses a significant challenge in identifying drivers who exceed the speed limit or break other traffic regulations. Thus, in our research, we evaluated the effectiveness of the suggested combination and other known AI methods in minimizing data loss, using Eq. (11).

Eq. (11) states that a data transmission strategy is considered efficient to boost data integrity if it has a reduced percentage of data loss. The efficacy of the proposed combination, as well as other AI approaches, in protecting against data loss has been calculated and presented in Figure 9.

Figure 9:

Further information about the elements of Figure 9 can be found in Table 8, which also emphasizes the effectiveness of the suggested integrated and other systems in terms of their ability to deliver a lower proportion of data loss.

Figure 9 and Table 8 illustrate that the suggested integrated strategy exhibits the lowest proportion of information loss compared to other relevant methodologies. Therefore, we may assert that enhancing data integrity is efficient. The efficacy of any error control strategy can be assessed using the signal-to-noise ratio (SNR) [16, 42]. An error control strategy that provides superior SNR values is deemed efficient in minimizing data errors and improving data integrity. The SNR can be computed using the subsequent Eq. (12).

In Eq. (12), i(n) represents the mean value of the output where n is any number. Here, the performance of the proposed integrated technique to reduce data error has been calculated in terms of SNR with the help of Eq. (12) and plotted in Figure 10. Consequently, the performances of some well-known error controlling approaches from the contemporary literature have been calculated in terms of their capacity to produce higher SNR values and plotted in Figure 10 further for drawing a comparison among them.

Figure 10:

Table 9, which also highlights the efficacy of the proposed integrated and other systems in terms of their capacity to produce a higher proportion of SNR Decibel (DB), provides more details regarding the components of Figure 10.

Figure 10 and Table 9 demonstrate that the suggested error control strategy provides a greater SNR value compared to existing error control methods. Consequently, it can mitigate data loss by reducing various transmission faults. Additionally, the efficacy of information regeneration in the proposed integrated methodology has been compared with that of established methodologies in the present literature, specifically regarding their percentage of packet loss estimated post-implementation. Literature indicates that the assessment of information loss percentage reveals the strengths and limitations of a transportation system, as well as the efficacy of error control techniques in mitigating information loss. The literature identifies several prominent methodologies, including LDPC, Reed-Solomon error correction, FEC, error correction and lost bit generation through Hamming distance computation, and Cyclic Redundancy Checking [44, 48]. Consequently, the percentage of packet loss has been computed for the suggested method and the current methods utilizing the following Eq. (9).

Per the convention, a data recovery method is deemed efficient if it results in a reduced percentage of packet loss or the opposite. Consequently, the efficacy of the suggested method and other current methods to diminish the packet loss percentage has been assessed using Eq. (13). The sample information size is determined by the capacity of the input storage device, ranging from 2 GB to 5 GB throughout the transmission procedure. The percentage of packet loss across all transmissions has been computed using Eq. (13) for all transmission phases; their average has been calculated and plotted in Figure 11.

Figure 11:

Table 10, which also highlights the efficacy of the proposed integrated and alternative systems in terms of their capacity to deliver a lower fraction of packet loss, provides more details regarding the components of Figure 11.

Figure 11 and Table 10 demonstrate that the proposed combination effectively reduces the proportion of packet loss compared to other existing techniques. Consequently, Figure 8, Table 7, Figure 9, Table 8, Figure 11, and Table 10 demonstrate that the proposed integrated technique effectively minimizes data loss by yielding a reduced proportion of packet loss and data loss while providing an enhanced SNR. It yields a reduced proportion of data loss, diminished packet loss, and an enhanced SNR relative to other established alternatives. Consequently, the proposed amalgamation can fulfill our second objective of this research.

c. Analysis of its time efficiency

The temporal efficiency of information processing is crucial for every information processing system. In our method, the primary concern is the processing time, as faster information processing can prevent the car driver from exceeding the speed limit or breaching any defined speed restrictions on the highway. The processing efficiency of an information processing system can be determined by calculating its ability to generate lower cyclomatic complexity (CC). The CC [16, 27, 42] can be used to quantify the efficiency of the system being utilized and its ability to solve certain tasks related to the source problem. Typically, the measurement involves creating a Control Flow Graph of the software module to determine the overall count of linearly independent pathways within the module [20]. The CC can be determined with the following Eq. (14).

The CC represents the number of edges on the control flow graph. The number of nodes in the graph is given by N, whereas T represents the number of linked components in the graph, and C represents the number of connected components. The table provided, Table 11, outlines the meanings associated with different ranges of CC.

Among the AI approaches we've covered so far, Table 11 shows that the proposed combination has the lowest CC. The proposed integrated method is exceptionally successful in reducing time complexity and processing information efficiently. The Processing Capacity [44, 48] can be quantified by its capability to manage a specific volume of information per second, enabling the assessment of the temporal efficiency of various AI systems. Utilizing Eq. (10), one can determine the computation of processing capacity as follows.

The convention dictates that a technique with high processing capacity is time-efficient, and conversely, a technique with low processing capacity is time-in-efficient. Thus, utilizing Eq. (15), we can calculate the information processing capability of the suggested combination alongside other pertinent existing AI methodologies, as illustrated in Figure 12.

Figure 12:

More information on the elements of Figure 12 can be found in Table 11, which also emphasizes the effectiveness of the suggested integrated and alternative systems in terms of their ability to offer higher information processing capacity.

The findings presented in Figure 12 and Table 12 demonstrate that the proposed combination provides a greater capacity for information processing than the other AI techniques that are already in use. Since this is the case, Figure 12, Tables 12 and 11 provide evidence that the proposed combination is significantly more effective and efficient in terms of processing the information when compared to the other AI systems that are currently in use. As a result, these discoveries provide support for the third purpose of this research.

d. Worst-case scenario discussion for the proposed technique

The intended outcome can be attained by placing Bluetooth devices at every 1 km interval along the route to capture vehicle information. Should any device malfunction under any circumstances, the alternative device can supply backup data with minimal error rates or without mistakes altogether. Conversely, communication delay is a significant concern in such circumstances. Consequently, we will evaluate the network robustness of our proposed design utilizing the information from references [6, 29, 42].

1. The used many-to-one Bluetooth device consists of eight communication channels, that is, it can communicate with a maximum of eight cars at a time, which is sufficient for the highways. The minimum communication rate of a many-to-one or one-to-many Bluetooth device is 800 kbps, which means each channel of a many-to-one Bluetooth device can communicate at 100 kbps.

2. As per the proposed model, each handshaking message size will be a maximum of 5 KB. Hence, the transportation of a 5 KB message will take a maximum of 0.05 s.

3. The range of the used Bluetooth device in our hypothetical architecture is 100 m.

4. An over-speeded car with 240 km/hr (maximum) can travel 3.3 m in 0.05 s.

If certain circumstances, the network speed comes down to 4 kbps, then it can also communicate with an overspeeded car which is running at 240 km/hr. Hence, in both cases, the proposed architecture can handle.