An Efficient Model for Vehicular Ad Hoc Networks using Machine Learning and High- Performance Computing

doi:10.21203/rs.3.rs-3457206/v1

Download PDF

Research Article

An Efficient Model for Vehicular Ad Hoc Networks using Machine Learning and High- Performance Computing

https://doi.org/10.21203/rs.3.rs-3457206/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 10 Jun, 2024

Read the published version in The Journal of Supercomputing →

You are reading this latest preprint version

Vehicular Ad Hoc Networks (VANETs) are recent advancement that permits vehicles to communicate with one another and with infrastructure, improving road safety and traffic efficiency. One of the difficulties in constructing and maintaining VANETs is dealing with the consequences of blockage, which can occur when buildings, trees, or other obstructions block radio signals between vehicles. However, the presence of vehicles as obstacles can severely impact the performance of VANETs. In this paper, an efficient machine learning (ML)-based technique is used to identify the impact of vehicle obstacles in VANETs. The proposed Tree-based models showed better results in comparison to the state-of-the-art models in all the tests conducted. The findings of the proposed model outperform the existing models and demonstrate that the proposed models can precisely predict and classify data, which makes it an important tool for various applications where accurate classification is crucial.

Machine Learning (ML)

VANETs

Vehicle Obstacle Impact

Optimizable Tree

Quality Metrics

High-Performance Computing

VANETs are nowadays becoming a center of attraction due to their capability for the improvement of safe transport, traffic efficiency, and passenger comfort [1]. The recent issue in VANETs is to prevent vehicle accidents caused by obstacles such as pedestrians, animals, and other vehicles. In order to solve these issues, various solutions have been given, which comprise ML based approaches [2]. ML used in VANETs for obstacle impact mitigation has gained momentum due to its learning capability from historical data and making accurate predictions in real-time. These methods analyses big datasets which identified patterns and trends that can predict obstacles [3, 4]. Nowadays, Deep Learning (DL) techniques have shown better outcomes for huge datasets that can predict obstacle impact [5]. These techniques have been developed for the transport safety and to minimize the number of accidents [6]. These methods have many advantages like improvement in accuracy and efficiency of obstacle detection and prediction, making the system more robust and reliable [7].

These techniques analyses real-time data from different sources like GPS, communication networks etc. [8].

We proposed a data-driven approach, the Optimizable Tree ML technique, for the improvement of accuracy and efficiency in detecting and predicting obstacles. Moreover, we evaluate the proposed approach using real-world data and demonstrate its effectiveness in mitigating vehicle obstacle impact in VANETs. Finally, the proposed approach is compared with existing approaches and shows that it outperforms them in terms of accuracy, precision, and scalability.

The remaining portion of this paper is arranged as given below. Section 2 elaborates the review of the literature on ML-based obstacle detection and prediction in VANETs. Section 3 deliberates the proposed approach in detail, including the data collection process, feature extraction, and the various ML techniques used. Section 4 presents the experimental setup and the findings then are discussed in Section 5. Lastly, the conclusion of the paper highlights future scope in Section 6.

The problem of blockage prediction in VANETs refers to the prediction and identification of potential obstacles or barriers that can disrupt wireless communication between vehicles. Blockages can be caused by different parameters like buildings, vegetation, and vehicle obstructions, resulting in non-line-of-sight (NLOS) conditions and significant signal attenuation. The research motivation behind addressing this problem lies in the critical need for reliable and efficient communication in VANETs. VANETs play a crucial role in enabling various applications such as traffic management, safety systems, and entertainment services. However, the presence of blockages can severely impact the performance and effectiveness of these applications. Therefore, accurately predicting and anticipating blockages can lead to improved communication reliability, enhanced safety, and efficient traffic management in VANETs. By developing effective blockage prediction techniques, researchers aim to overcome the limitations posed by NLOS conditions and enable proactive measures to mitigate the effects of blockages. Predicting blockages can help vehicles make informed routing decisions, optimize resource allocation, and improve communication protocols to maintain seamless connectivity. Additionally, accurate blockage prediction can assist in developing Intelligent Transportation Systems (ITS) and enabling Advanced Driver Assistance Systems (ADAS) that rely on real-time and uninterrupted communication between vehicles. The authors in [9] measured and parameterized multi-band (6.75 GHz, 30 GHz, 60 GHz, and 73 GHz) V2V channels in urban and highway settings in Germany and China, focusing on the effects of vehicle obstruction. The authors of this study highlight how ML techniques might be implemented for VANETs to function effectively. The study is evaluated from a variety of angles, including traffic, communication, and safety [10]. Even in the sub-6-GHz frequency band, the effect of obstacles on V2V channel characteristics has not been adequately addressed, and it remains an outstanding topic in the mmWave region. NLOS occurs in real-world V2V conditions because of blockades produced by various vehicle sizes, buildings, and vegetation. NLOS occurs commonly for mmWave V2V channels because of traditional path loss [11]. Traffic management, entertainment, and traffic safety are just a few of the many possible uses for VANETs. In [12] beam-based approach the result is heavily reliant on traffic and network obstructions. AI may be utilized to successfully overcome problems in traditional vehicular communications-based applications [13]. Then the non-parametric and data-driven approach based on an enhanced k-NN technique has shown promise for short-term traffic prediction, particularly in cases where historical data is available and no significant changes in road infrastructure or traffic patterns are proposed [14]. However, the paper [15] proposed a model based on optimizable k-nearest neighbors (k-NN) that has shown promise for VANET. Afterward, in [16] the VANET based on adaptive neighborhood selection using an expansion strategy has shown promise for the VANET. There are several suggested methods for urban traffic flow prediction, each utilizing different techniques to achieve high accuracy. ANN-based Gaussian mixture clustering model is used to predict traffic flow [17]. In [18] hybrid predictive framework based on SVR and enhanced GA method uses an SVR algorithm to identify optimal parameters for forecasting models and an advanced GA with chaotic characteristics. This hybrid framework has improved prediction accuracy and reduced computational time. In [19] ML algorithm utilizes multiple sources of data, including historical traffic flow data and real-time weather data, and multiple measures of traffic flow, such as volume, speed, and density. Multiple LSSVR models utilize multiple LSSVR models, each with a different time lag and performance. The models are based on Gaussian kernel functions and are trained on historical traffic flow data [20]. To best route data packets to their destination in vehicular communication, the authors of [21] presented an Intelligent Beaconless (IB) geographical forwarding technique. For GPS outage issues in environments using vehicular cyber physical systems (VCPS), a geometry-based location estimate method is provided in [22].

The research gap identified in the given information is that even in the sub-6-GHz spectrum, the effect of obstacles on V2V channel characteristics has not been adequately addressed. Additionally, different researches have introduced approaches to enhance prediction capabilities in urban traffic scenarios. A data driven model has been developed to predict VANETs traffic pattern which improves accuracy.

VANETs is a transportation network in which obstacles may disturb the communication. However, several challenges can hamper this such as obstacle can blocks communication. These techniques can develop effective collision avoidance systems which reduces the risk of accidents [23]. These issues can address by measuring different frequency bands of V2V channels. Development of data-driven methods can play a significant role in tackling this problem. Mathematically Blockage Prediction problem in VANETs represented by a set of vehicles $V=\{{v}_{1}, {v}_{2},\dots \dots \dots ,{v}_{n}\}$ moves in a geographic area $Ar$. Every vehicle ${v}_{i}$ has communication devices. The main objective is to detect the significance of obstacles on V2V communication channels, Identification of blockages caused by obstacles such as buildings, vegetation, and vehicle sizes. Optimization of management strategies of traffic in VANETs.The problem formulation is as given

1. Variables:

Binary variable $B(v, t)$: Indicates the presence of blockage for vehicle $v$ at time $t$. $B\left(v, t\right)=1$ if there is a blockage, and $B\left(v, t\right)=0$ otherwise.

2. Parameters:

Position $P(v, t)$:The geographical position of vehicle $v$ at time $t$.
Communication signal strength $S(v, t)$: The strength of the V2V communication signal for vehicle $v$ at time $t$.
Obstacle characteristics $O(v, t)$: Descriptive parameters representing the type, size, and location of obstacles in the vicinity of vehicle $v$ at time $t$.

3. Constraints:

Line-of-sight (LOS) constraint: The LOS condition between two vehicles ${v}_{1}$ and ${v}_{2}$ must be satisfied for successful communication, i.e., $S({v}_{1} , t)>Threshold$ and $S({v}_{2} , t)>Threshold$, where the threshold represents the minimum required signal strength for reliable communication.

4. Objective:

Predicting blockages: Determine the values of $B(v, t)$ for all vehicles $v$ and time intervals $t$, that indicate the presence or absence of blockages in V2V communication channels.

The impact of vehicles as obstacles in VANETs can be modeled mathematically using different performance metrics.

When an obstacle blocks or interferes with communication between two vehicles, the packet delivery ratio may decrease, which can lead to packet’s high probability loss and delay [24]. This can be expressed mathematically as:

$$Packet Delivery Ratio =\frac{Number of Packets Successfully Delivered}{Total Number of Packets Sent} \left(1\right)$$

Similarly, the cumulative delay is derived as the time required for a packet transmitted from the sender to the receiver. When an obstacle blocks or interferes with communication, the cumulative delay may increase, which can affect the performance of different applications, like collision avoidance or emergency warning systems. This can be expressed mathematically as:

$$End to End Delay = Time taken for Packet to be Transmitted from Source to Destination \left(2\right)$$

Finally, the network throughput is derived as the amount of data transmitted per unit of time. When obstacles block or interfere with communication, the network throughput may decrease, which can result in reduced traffic flow and increased travel time. This can be expressed mathematically as:

$$Network Throughput = Amount of Data Transmitted/Unit Time \left(3\right)$$

The system model for probable blockages in VANETs given in Fig. 1 refers to the conceptual framework that captures the factors and components involved in identifying potential blockages or obstacles in the context of VANETs. In this framework 1, 2, 3, and 4 shows the blockages or obstacles and V2V communications are represented by 5. It has vehicles which has on-board units (OBUs)involved in sharing data, road network that operates vehicles, VANET which facilitates vehicle-to-infrastructure (V2I) and V2V communication, sensor data for detecting and gathering data, data fusion model used for exchanging information that improves the accurate and reliable blockage detection, blockage detection is used to detect potential blockages or obstacles on the road. Other than that, it has alert generation model which generates alerts or warnings to inform the affected vehicles or the surrounding infrastructure. Based on this model it has decision-making system it reroutes the vehicles, adjusted traffic signal timings, or notifying emergency services.

An Optimizable Tree is a DT-based technique which is designed to minimize a particular objective function. It optimizes gradient-based optimization technique like stochastic gradient descent and has several advantages over state of art like flexibility for capturing more complicated decision planes and optimizes over fitting problem. The Optimizable Tree given by training dataset express in Eq. (4) and split function Eq. (5).

$$D=\left\{\left({x}_{1}, {y}_{1}\right),\left({x}_{2}, {y}_{2}\right) , \dots \dots \dots , \left({x}_{N}, {y}_{N}\right) \right\}$$

The split function $s\left(.\right)$ that takes an input feature vector $x$ and returns a binary decision, i.e., $s\left(x\right) ϵ \left\{0, 1\right\}$. If $s\left(x\right)=1$, then the example $x$ is the left child node of the current tree node, and if $s\left(x\right)=0$, then it is the right child node.

The split function $s\left(.\right)$ can be expressed by a weight vector $w ϵ {\mathbb{R}}^{d}$ and a bias term $b ϵ \mathbb{R}$, i.e.,

$$s\left(x\right)=\left\{1 if wT\left(x\right) + b > 0, 0 otherwise\right\}$$

The weight vector and bias term are learned during training by optimizing a given objective function. The key function is typically defined as a sum of per-example loss functions

$$\mathcal{l}=\left({y}_{i }, {T}_{i}\left({x}_{i }\right)\right)$$

that measure the difference between the predicted label ${T}_{i}\left({x}_{i }\right)$ and the true label ${y}_{i }$, i.e.,

$$F\left(w, b\right)=\frac{1}{N}\sum _{i=1}^{N}\mathcal{l}=\left({y}_{i }, {T}_{i}\left({x}_{i }\right)\right)$$

The per-example loss function $\mathcal{l}$ can be defined based on the specific task.

During training, the weight vector and bias term are updated using gradient descent or a variant thereof, i.e.,

$$w\leftarrow w-\eta {\nabla }_{w}\left(F\left(w, b\right)\right)$$

and

$$b\leftarrow b-\eta {\nabla }_{b}\left(F\left(w, b\right)\right)$$

where η is the learning rate and ${\nabla }_{w}$ and ${\nabla }_{b}$ are the gradients of the w and b, correspondingly.

The training process continues until convergence or a predefined stopping criterion is met. The resulting decision tree $T$ is then used to make predictions on new, unseen examples by traversing the tree and applying the learned decision rules to the input feature vector.

An efficient Machine Learning-Based Approach for predicting Vehicle Obstacle Impact in VANETs using Optimizable Trees:

Collect data: Collect sensor data from vehicles, such as GPS location, speed, and acceleration, as well as obstacle data, such as location, size, and type.

Data pre-processing: Pre-process the data by removal of duplicates and normalizing the numerical features.

Feature extraction: Extract the features related from the pre-processed data, for instance distance to the nearest obstacle, angle of the obstacle relative to the vehicle's heading, and the vehicle's speed and acceleration.

Split data: Divide the dataset into training and testing sets.

Train the Optimizable Tree: Train an Optimizable Tree model on the extracted features and respective labels, such as whether a collision occurred or not, using gradient-based optimization.

Hyperparameter tuning: Optimize the hyperparameters of the Optimizable Tree model, using a grid search or other optimization method.

Model evaluation: Assess the proposed approach on the testing set, using suitable evaluation metrics.

Model deployment: Deploy the trained model in a real-world scenario, such as in a VANETs, to predict the likelihood of collisions with obstacles.

Model maintenance: Update and retrain the model periodically as new data becomes available, and continuously evaluate and monitor its performance to ensure reliable collision prediction in real-time scenarios.

The Optimizable Tree model takes a set of features$X=\left\{{x}_{1}, {x}_{2}, \dots \dots \dots , {x}_{n} \right\}$ as input and produces a binary classification output y, which indicates whether a collision with an obstacle is likely or not.

The output of the Optimizable Tree model is defined as follows:

$$y=\text{s}\text{i}\text{g}\text{m}\text{o}\text{i}\text{d}\left(\text{z}\right)$$

where sigmoid is the logistic sigmoid function and z is the output of the final node in the tree which is defined as:

$$z= {w}^{T}h\left(x\right)$$

where ${w}^{T}$ denotes the transpose operation of weight vector and $h\left(x\right)$ is a vector of features computed from the input features $x$.

The features $h\left(x\right)$ are computed by recursively applying a set of decision functions ${f}_{i}\left(x\right)$ to the input features $x$, where i ranges from 1 to the number of nodes in the tree:

$$h\left(x\right)=\left[{f}_{1}\left(x\right), {f}_{2}\left(x\right), \dots \dots \dots \dots , {f}_{m}\left(x\right)\right]$$

where, $m$ is the number of nodes in the tree.

Each decision function ${f}_{i}\left(x\right)$ takes the form:

$${f}_{i}\left(x\right)=ReLU\left({\left({w}_{i}\right)}^{T}{x}_{i}+{b}_{i}\right)$$

where $ReLU$ is the rectified linear unit function, ${w}_{i}$ is a weight vector, ${b}_{i}$ is a bias term, and x is the input feature vector.

The weight vectors and bias terms are learned during training using gradient-based optimization, where the objective is to minimize a binary cross-entropy loss function:

$$L\left(y, t\right)=-t\text{log}y-\left(1-t\right)\text{log}\left(1-y\right)$$

Where, $t$ is the ground-truth label (1 for collision and 0 for non-collision).

The gradient loss weight vectors and bias terms is computed using backpropagation, and the weights are updated using a gradient descent algorithm, such as Adam or SGD.

During hyperparameter tuning, the rate of learning and the number of layers can be varied to find the optimal combination of hyperparameters that maximizes the performance of the model on the validation set.

The confusion matrix (CM) is used to assess the performance of an ML technique. The given parameters are required in data-driven models for comparative analysis.

	Positive (Predicted)	Negative (Predicted)
Positive (Actual)	${\text{T}}_{\text{P}}$	${\text{F}}_{\text{N}}$
Negative (Actual)	${\text{F}}_{\text{P}}$	${\text{T}}_{\text{N}}$

In this, the classifier correctly predicted ${\text{T}}_{\text{P}}$ and ${\text{T}}_{\text{N}}$ instances and incorrectly predicted

and ${\text{F}}_{\text{N}}$ instances. The CM can be used for calculating different performance metrics like $A$, $P$, $R$, $F$, etc.

Accuracy $\left(A\right)$: Accuracy is a measure of how close a measured value (MV) is to the true value (TV) of the quantity being measured. It is computed by using Eq. (15).

$$A=\left(\frac{{\text{T}}_{\text{P}}+ {\text{T}}_{\text{N}}}{{\text{T}}_{\text{P}}+ {\text{F}}_{\text{P}}+ {\text{T}}_{\text{N}}+{\text{F}}_{\text{N}} }\right)$$

Precision $\left(P\right)$: Eq. (16) define the precision (P) which depicts how closely the MV associated with one another.

$$P=\left(\frac{{\text{T}}_{\text{P}}}{{\text{T}}_{\text{P}}+ {\text{F}}_{\text{P}}}\right)$$

Recall $\left(R\right)$: It is ability of technique to find all relevant things within a dataset given by Eq. (17).

$$R=\left(\frac{{\text{T}}_{\text{P}}}{{\text{T}}_{\text{P}}+{\text{F}}_{\text{N}}}\right)$$

F-Score $\left(F\right)$: It can be computed as the harmonic average of R and P given in Eq. (18).

$$F=\left(\frac{2PR}{P+R}\right)$$

Specificity $\left(S\right)$: It is given in Eq. (19).

$$S=\left(\frac{{\text{T}}_{\text{N}}}{{\text{F}}_{\text{P}}+ {\text{T}}_{\text{N}}}\right)$$

Error Rate $\left(ER\right)$: It is computed by Eq. (20)

$$ER=\left(\frac{No. of incorrect predictions}{Size of the dataset}\right)$$

In this comparison given in Table 1 of ML models: Linear SVM: Moderate accuracy (66.1% training, 66% testing), takes longer training and testing time, and incurs higher costs. Kernel Naive Bayes: Reasonable accuracy (76.3% training, 77% testing), faster training and testing, and lower costs. Neural Network: High accuracy (89.4% training, 91.5% testing), fast training, but specific costs not provided. Decision Tree: Excellent accuracy (98.27% training, 100% testing), extremely fast training, and minimal costs., 84.3% testing). Optimizable Tree: Near-perfect accuracy (99.9% training, 100% testing), very fast training, and lowest costs. The "Optimizable Tree" model appear to offer the best trade-offs between accuracy, training speed, and costs.

The time complexity (TC) of the Optimizable Tree algorithm depends on several factors, such as the number of input features, the depth and width of the decision tree, and the number of training iterations. Assuming there are $n$ input features and m nodes in the decision tree, the TC of computing the output for a single instance is $O\left(mn\right)$. This is because each node in the tree computes a linear function of the input features, which involves a dot product between the weight vector and the input feature vector. During training, the computational complexity of computing the gradients and updating the weights iterations and the batch size. Assuming there are $N$ training instances and the batch size is $B$, the TC of a single training iteration is $O\left(BNmn\right)$, since we need to compute the output and gradients for each instance in the batch. The total TC of proposed model depends on the number of training iterations, which is typically several hundred to several thousand, and the batch size, which is typically several tens to several hundred. Therefore, the overall TC is typically $O\left(BNTmn\right)$, where $T$ is the number of training iterations and B is the batch size. The space complexity of the Optimizable Tree algorithm depends on the number of nodes in the decision tree and the number of parameters in the model, which is proportional to the number of nodes. Therefore, the space complexity is typically $O\left(mn\right)$. During training, additional memory is required to store different parameters, which is proportional to the batch size. Therefore, the space complexity during training is $O\left(Bmn\right)$.

Table 1

Comparison table for the performance of existing and proposed ML model.
Model	Total Cost (Training in sec)	Total Cost (Testing in sec)	Prediction speed in obs/sec. (Training)	Training Time in sec.	Training Accuracy %	Testing Accuracy %
Linear SVM [13, 27, 30, 31]	7133	17866	310000	403.43	66.1	66
Neural Network [25, 26, 28, 29]	NA	NA	1100000	29.256	89.4	91.5
Kernel Naive Bayes [31]	4996	12090	460	139.73	76.3	77
Decision Tree [32]	359	21	1500000	1.0016	98.2677	100
Optimizable Tree	27	2	2500000	26.441	99.9	100

Table 2 shows the performance metrics of decision tree and proposed ML algorithms for obstacle impact in VANETs which can also visualize in Fig. 2. The bar graph's main purpose is to provide a visual overview of the performance metrics of decision tree and proposed model. The actual bar lengths and proportions would depend on the exact values. By using confusion matrixes of the decision Tree, which achieves a high sensitivity of 98.425% and a high specificity of 98.1104%. It also has a high balanced accuracy of 98.2677% and a high accuracy of 98.3%. However, its error rate is 1.7%, which is relatively high. The precision and F1-score of Fine Tree are also high, with values of 98.75% and 98.5872%, correspondingly. The proposed model Optimizable Tree, achieves a very high sensitivity, specificity, balanced accuracy, and accuracy of 99.9%. Its error rate is only 0.1%, making it the most accurate algorithm. The precision and F1-score of Optimizable Tree are also very high, with values of 99.9%.

Hence, the Optimizable Tree algorithm appears to be the most effective in predicting vehicle obstacle impact in VANETs, achieving the highest accuracy, precision, and F1-score while also having a very low error rate. Thus, it is important to take other parameter like time complexity and implementation feasibility when choosing a machine learning algorithm for a real-world application.

Table 2

Performance Matrices of Decision tree and Proposed Optimizable tree ML techniques.
Model classifier	Sensitivity or Recall or TPR %	Specificity or TNR %	BA %	Accuracy %	Error rate %	Precision %	F1-Score %	FNR %	FPR %
Decision Tree [32]	98.425	98.1104	98.2677	98.3	1.7	98.75	98.5872	1.575	1.8895
Optimizable Tree	99.9	99.9	99.9	99.9	0.1	99.9	99.9	0.1	0.1

Figure 2 also depicts the performance metrics of decision tree and proposed ML algorithms for obstacle impact in VANETs. The bar graph's main purpose is to provide a visual overview of the performance metrics of decision tree and proposed model. The actual bar lengths and proportions would depend on the exact values.

The problem of blockage prediction in VANETs involves developing models and techniques to accurately predict and anticipate instances where radio signals between vehicles and roadside infrastructure are obstructed. The key goal is to improve the performance, safety, and efficiency of VANETs by identifying potential blockages caused by buildings, trees, or other obstacles, and modifying their impact on communication and connectivity. In conclusion, the paper presents ML based approach to predict the impact of vehicle obstacles in VANETs. The proposed Optimizable Tree model outperformed state-of-the-art models in various tests, achieving high scores in terms of several performance metrics. This indicates that the proposed model can accurately predict and classify data, which is essential for applications where accurate classification is crucial. The future scope for blockage prediction in VANETs includes refining data collection and feature engineering approaches to enhance the accuracy and generalizability of prediction models.

Conflict of Interest: No.

Funding: No

Financial Interest: No.

Availability of Data and Code: Data and code will be available on behalf of a request from the corresponding author.

Al-Sultan, S., Al-Doori, M. M., Al-Bayatti, A. H., & Zedan, H. (2014). A comprehensive survey on vehicular ad hoc network. Journal of network and computer applications, 37, 380-392.
Abdi, L., & Meddeb, A. (2018). Driver information system: a combination of augmented reality, deep learning and vehicular Ad-hoc networks. Multimedia Tools and Applications, 77, 14673-14703.
Iqbal, R., Doctor, F., More, B., Mahmud, S., & Yousuf, U. (2020). Big data analytics: Computational intelligence techniques and application areas. Technological Forecasting and Social Change, 153, 119253.
Olugbade, S., Ojo, S., Imoize, A. L., Isabona, J., & Alaba, M. O. (2022). A Review of Artificial Intelligence and Machine Learning for Incident Detectors in Road Transport Systems. Mathematical and Computational Applications, 27(5), 77.
Mchergui, A., Moulahi, T., & Zeadally, S. (2022). Survey on artificial intelligence (AI) techniques for vehicular ad-hoc networks (VANETs). Vehicular Communications, 34, 100403.
Luo, L., Sheng, L., Yu, H., & Sun, G. (2021). Intersection-based V2X routing via reinforcement learning in vehicular ad hoc networks. IEEE Transactions on Intelligent Transportation Systems, 23(6), 5446-5459.
Khatri, S., Vachhani, H., Shah, S., Bhatia, J., Chaturvedi, M., Tanwar, S., & Kumar, N. (2021). Machine learning models and techniques for VANET based traffic management: Implementation issues and challenges. Peer-to-Peer Networking and Applications, 14, 1778-1805.
Arif, M., Wang, G., Bhuiyan, M. Z. A., Wang, T., & Chen, J. (2019). A survey on security attacks in VANETs: Communication, applications and challenges. Vehicular Communications, 19, 100179.
M. Boban, D. Dupleich, N. Iqbal, J. Luo, C. Schneider, R. Müller, Z. Yu, D. Steer, T. Jämsä, J. Li, R. S. Thomä: 'Multi-Band Vehicle-to-Vehicle Channel Characterization in the Presence of Vehicle Blockage', IEEE Access, 2019.
Khatri, S., Vachhani, H., Shah, S. et al. Machine learning models and techniques for VANET based traffic management: Implementation issues and challenges. Peer-to-Peer Netw. Appl. 14, 1778–1805 (2021).
R. He, B. Ai, G. Wang, Z. Zhong, C. Schneider, D. A. Dupleich, R. S. Thomä, M. Boban, J. Luo, and Y. Zhang: 'Propagation Channels of 5G Millimeter Wave Vehicle-to-Vehicle Communications: Recent Advances and Future Challenges', IEEE Vehicular Technology Magazine, 2019.
G. H. Sim, S. Klos, A. Asadi, A. Klein and M. Hollick, "An Online Context-Aware Machine Learning Algorithm for 5G mmWave Vehicular Communications," in IEEE/ACM Transactions on Networking, vol. 26, no. 6, pp. 2487-2500, Dec. 2018, doi: 10.1109/TNET.2018.2869244.
W. Tong, A. Hussain, W. X. Bo and S. Maharjan, "Artificial Intelligence for Vehicle-to-Everything: A Survey," in IEEE Access, vol. 7, pp. 10823-10843, 2019, doi: 10.1109/ACCESS.2019.2891073.
Yu, Bin, Xiaolin Song, Feng Guan, Zhiming Yang, and Baozhen Yao. "k-Nearest neighbor model for multiple-time-step prediction of short-term traffic condition." Journal of Transportation Engineering 142, no. 6 (2016): 04016018.
Su, Ziyi, Qingchao Liu, Jian Lu, Yingfeng Cai, Haobin Jiang, and Lukuman Wahab. "Short-time traffic state forecasting using adaptive neighborhood selection based on expansion strategy." Ieee Access 6 (2018): 48210-48223.
Liu, Qingchao, Yingfeng Cai, Haobin Jiang, Xiaobo Chen, and Jian Lu. "Traffic state spatial-temporal characteristic analysis and short-term forecasting based on manifold similarity." Ieee Access 6 (2018): 9690-9702.
Zhao, Liang, Yufei Liu, Ahmed Y. Al-Dubai, Albert Y. Zomaya, Geyong Min, and Ammar Hawbani. "A novel generation-adversarial-network-based vehicle trajectory prediction method for intelligent vehicular networks." IEEE Internet of Things Journal 8, no. 3 (2020): 2066-2077.
Peng, Shanbi, Zhe Zhang, Enbin Liu, Wei Liu, and WeibiaoQiao. "A new hybrid algorithm model for prediction of internal corrosion rate of multiphase pipeline." Journal of Natural Gas Science and Engineering 85 (2021): 103716.
Cheng, Anyu, Xiao Jiang, Yongfu Li, Chao Zhang, and Hao Zhu. "Multiple sources and multiple measures-based traffic flow prediction using the chaos theory and support vector regression method." Physica A: Statistical Mechanics and its Applications 466 (2017): 422-434.
Alabi, Tobi Michael, Emmanuel I. Aghimien, Favour D. Agbajor, Zaiyue Yang, Lin Lu, Adebusola R. Adeoye, and Bhushan Gopaluni. "A review on the integrated optimization techniques and machine learning approaches for modeling, prediction, and decision making on integrated energy systems." Renewable Energy (2022).
Ghafoor, K.Z., Abu Bakar, K., Lloret, J. et al. Intelligent beaconless geographical forwarding for urban vehicular environments. Wireless Netw 19, 345–362 (2013).
Kaiwartya, O., Cao, Y., Lloret, J., Kumar, S., Aslam, N., Kharel, R., ... & Shah, R. R. (2018). Geometry-based localization for GPS outage in vehicular cyber physical systems. IEEE Transactions on Vehicular Technology, 67(5), 3800-3812.
Mchergui, A., Moulahi, T., & Zeadally, S. (2022). Survey on artificial intelligence (AI) techniques for vehicular ad-hoc networks (VANETs). Vehicular Communications, 34, 100403.
Gao, H., Liu, C., Li, Y., & Yang, X. (2020). V2VR: reliable hybrid-network-oriented V2V data transmission and routing considering RSUs and connectivity probability. IEEE Transactions on Intelligent Transportation Systems, 22(6), 3533-3546.
Hossain, M. A., Noor, R. M., Yau, K. L. A., Azzuhri, S. R., Z’aba, M. R., & Ahmedy, I. (2020). Comprehensive survey of machine learning approaches in cognitive radio-based vehicular ad hoc networks. IEEE Access, 8, 78054-78108.
Jensen, M. B., Philipsen, M. P., Møgelmose, A., Moeslund, T. B., & Trivedi, M. M. (2016). Vision for looking at traffic lights: Issues, survey, and perspectives. IEEE transactions on intelligent transportation systems, 17(7), 1800-1815.
Ye, Y. Y., Hao, X. L., & Chen, H. J. (2018). Lane detection method based on lane structural analysis and CNNs. IET Intelligent Transport Systems, 12(6), 513-520.
Kim, J., Kim, J., Jang, G. J., & Lee, M. (2017). Fast learning method for convolutional neural networks using extreme learning machine and its application to lane detection. Neural Networks, 87, 109-121.
Yuan, Y., Xiong, Z., & Wang, Q. (2016). An incremental framework for video-based traffic sign detection, tracking, and recognition. IEEE Transactions on Intelligent Transportation Systems, 18(7), 1918-1929.
Sun, J., & Sun, J. (2016). Real‐time crash prediction on urban expressways: identification of key variables and a hybrid support vector machine model. IET intelligent transport systems, 10(5), 331-337.
Nguyen, H., Cai, C., & Chen, F. (2017). Automatic classification of traffic incident's severity using machine learning approaches. IET Intelligent Transport Systems, 11(10), 615-623.
Soltani, S., & Mutka, M. W. (2015). A decision tree cognitive routing scheme for cognitive radio mesh networks. Wireless Communications and Mobile Computing, 15(10), 1405-1417.

No competing interests reported.

Download PDF

Journal Publication

published 10 Jun, 2024

Read the published version in The Journal of Supercomputing →

Editorial decision: Revision requested
09 May, 2024
Reviews received at journal
09 May, 2024
Reviews received at journal
09 May, 2024
Reviews received at journal
29 Apr, 2024
Reviewers agreed at journal
21 Apr, 2024
Reviewers agreed at journal
21 Apr, 2024
Reviewers agreed at journal
20 Apr, 2024
Reviewers agreed at journal
19 Apr, 2024
Reviewers agreed at journal
19 Apr, 2024
Reviewers agreed at journal
19 Apr, 2024
Reviewers invited by journal
19 Apr, 2024
Submission checks completed at journal
17 Oct, 2023
Editor assigned by journal
17 Oct, 2023
First submitted to journal
17 Oct, 2023

You are reading this latest preprint version

An Efficient Model for Vehicular Ad Hoc Networks using Machine Learning and High- Performance Computing

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Literature Review and Related Work

3. Proposed Model and Methodology

1. Variables:

2. Parameters:

3. Constraints:

4. Objective:

4. Results and Discussion

5. Conclusion and Future Work

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1

	Positive (Predicted)	Negative (Predicted)
Positive (Actual)	\({\text{T}}_{\text{P}}\)	\({\text{F}}_{\text{N}}\)
Negative (Actual)	\({\text{F}}_{\text{P}}\)	\({\text{T}}_{\text{N}}\)