Reliability is a critical issue in vehicular networks. A Deep Learning (DL) method is proposed in this study to automatically predict the reliability of Cognitive Radio Vehicular Networks (CR-VANETs) ignored in the previous research. First, a dataset is generated based on a previously proposed method for the reliability assessment of CR-VANETs. Then, a model is proposed to predict the networks' reliability using the DL method and compared with other machine learning methods. While machine learning methods have been applied in vehicular networks, they have not been used for reliability prediction. The proposed DL model is utilized in this research to predict CR-VANETs' reliability. Based on the results, the DL model outperforms other machine learning methods for reliability prediction. The correlation coefficient and root mean square error of the test data for the DL model are 0.9862 and 0.0381. These results indicate the CR-VANETs' reliability prediction accurately using the proposed method.
Research Article
A Deep Learning Method for Automatic Reliability Prediction of CR-VANETs
https://doi.org/10.21203/rs.3.rs-2604220/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Reliability Prediction
Dataset Generation
CR-VANETs
Deep Learning
Machine Learning
System reliability refers to the ability of a system's components to perform correctly for a given period under specific conditions [1]. Estimating the reliability of complicated systems is often challenging due to their multiple components, the functional interdependence of these components, and complex topologies. A system's reliability assessments require applicable analytic techniques before the commencement [2]. A system's reliability assessment mainly aims to quantify the behavior of the system related to its failures [3]. Such an assessment identifies and diagnoses system problems, thus improving the system's behavior and avoiding future problems.
One of the main requirements of Vehicular Ad-hoc NETwork (VANET) technology is to ensure road safety [4], as the reliability of communications is an essential issue in this application. Moreover, reliability estimations are crucial to evaluating and improving the dependability of vehicular safety messaging. A fast and efficient exchange of messages is necessary for V2V communications to enhance road safety. One of the main challenges associated with this requirement is achieving high reliability [5]. This challenge is caused by the fast movement of vehicular nodes and the instability of networks, which result in frequent changes in network topology [6]. Therefore, delays in message distribution and reliable communication for emergency messages are critical issues in VANETs [7]. The transmission of safety messages may become congested due to flooding, which reduces reliability [8]. The challenges associated with VANETs' communications are: 1) meeting safety system requirements and 2) providing reliable packet delivery with low communication latency [9] to ensure the timely and appropriate reactions of drivers.
During peak traffic hours, the assigned spectrum resources might be inadequate. Providing efficient access to more spectrum resources in underused bands is effective for transmission congestion reduction. This approach is applied in Cognitive Radio (CR) technology. Hence, the application of the CR in VANETs makes the CR-enabled vehicles detect spectrum holes for VANET communications, contributing to the timely broadcasting of safety messages [10]. The main motivations for utilizing CR technology are its permanency in highly reliable communications and its efficient application of spectrum resources [11]. CR technology improves the performance of vehicular communications providing efficient use of available bandwidth [12]. However, this technology is not flawless. Therefore, the assessment of CR-VANETs' reliability is essential to improving dependability. Moreover, considering drivers as one of the vehicles' components affects the reliability and safety of Intelligent Transportation Systems (ITSs). Hence, evaluating systems' reliability using automatic methods instead of real-world experiments or simulations saves time.
Machine learning has provided different tools for VANETs to generate, communicate, and share data [13]. However, a few studies have estimated systems' reliability using machine learning. Machine learning can handle incomplete data and perform predictions and generalizations after the training. A dataset is generated in this study using a previously proposed method [14]. To the best of our knowledge, the dataset created in this paper is the first of its kind. Then, a DL model is utilized to automate the CR-VANETs' reliability estimation and compare the results with Linear Regression (LR), Support Vector Machines (SVMs), Random Forest (RF), and bagging methods. DL could be helpful for reliability estimation due to its capability for complex pattern capture. The model's precision is demonstrated using the correlation coefficient (R), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) parameters. The advantages and contributions of this study are three-fold:
-
Previous works on DL and machine learning applications in VANETs are extensively reviewed.
-
A dataset is generated for the CR-VANETs with drivers using a previously proposed analytical method.
-
A DL model is proposed to automatically estimate the reliability of the CR-VANETs and compare it with different machine learning methods applied to the generated dataset.
This paper is organized as follows. The related works are surveyed in Section 2. The proposed DL model to predict the CR-VANETs' reliability is presented in Section 3. Section 4 provides the results and discussions, and Section 5 includes the conclusion and a suggestion for the future.
Although machine learning methods are not extensively utilized to predict the reliability of vehicular communications, there are many studies on other applications of these methods in VANETs. Some of the applied DL and machine learning methods are reviewed in this section.
A traffic flow prediction approach was presented in [15] based on DL; this method considered spatial and temporal correlations. The researchers determined traffic flow features using stacked auto-encoders. Moreover, a greedy layer-wise unsupervised learning algorithm was applied to improve the prediction performance through fine-tuning.
The proposed intrusion detection solution in [16] used Deep Neural Networks (DNNs). The training for feature extraction was performed using in-vehicle network packets to detect normal and hacking packets.
In [17], the authors proposed an ANN-based clustering algorithm based on the mobility and reliability of vehicles. The reliability, speed, and distance between the vehicles were considered for clustering, which increased stability and connectivity while decreasing bandwidth and delay.
In [18], the authors used a Deep Q-Network (DQN) to learn from the characteristics of the vehicles located in a roadside unit range to provide road safety, an acceptable level of Quality of Service (QoS), and lengthen the roadside unit's battery lifetime.
A method for misbehavior detection using ANN models was provided in [19]. The authors derived features showing the status of misbehavior, the environment, and communication to detect misbehavior data, and feed-forward and back-propagation algorithms were used for misbehavior classification.
A stereo vision-based method was presented in [20] for obstacle detection in VANETs in an urban environment. The utilized model was a Deep-Stacked Auto-encoder (DSA), with the k-nearest neighbor classifier using three real-life datasets.
The proposed DQN model in [21] handled speed, overtaking, and lane change decisions by training the vehicles.
A resource allocation mechanism for vehicle-to-vehicle communications was presented in [22]. This method was based on Deep Reinforcement Learning (DRL) to find the optimal spectrum and power for message transmissions between vehicles.
In [23], the authors studied the application of the Convolutional Neural Network (CNN) model for traffic control and optimal route assignment for vehicular communication. The proposed model was evaluated using various parameters, including delay, throughput, packet delivery ratio, and network load.
A clustering-based reliable routing was presented in [24]. In the proposed solution, clustering was performed using the Simulated Annealing (SA) algorithm, and a neural network was used to select the cluster's head.
A machine learning-based method for driving habit prediction was presented in [25] for stable clustering. The authors classified vehicles into two alignments using a naive Bayes classifier with different factors, such as relative speed, vehicle type, and the number of traffic violations. When combined with clustering design, the method was effective for stable clustering.
In [26], the authors found the optimum size of the Contention Window (CW) in VANETs using three Artificial Neural Network (ANN) algorithms, namely the particle swarm optimization, differential evolution algorithm, and artificial bee colony algorithm.
The method proposed in [27] was a hybrid SVM-based method for detecting Distributed Denial-of-Service (DDoS) attacks on VANETs. The SVM methods considered features like collisions, packet drop, and jitter under normal conditions and DDoS attacks.
Since network traffic prediction allows Intelligent Transport Systems (ITS) for proactive response to events before they happen, an LSTM DL model was purposed in [28] for VANETs traffic prediction. They evaluated the model based on RMSE and Mean Absolute Percentage Error (MAPE).
Intrusion detection was performed in [29] using a combination of auto-encoder networks and Recurrent Neural Networks (RNN).
The method described in [30] was also an Intrusion Detection System (IDS) for VANETs using DL, in which the researchers used the Deep Belief Network (DBN) algorithm for learning.
A machine learning-based hybrid intrusion detection model in VANETs was presented in [31]. The model had two components: it detected known attacks using RF classification, and it filtered dishonest nodes using coresets-based clustering.
An optimized method for VANET clustering inspired by whales' nature was described in [32]. This method optimized the number of clusters, hence, resource requirements reduction. Also, it increased the cluster lifetime.
The framework proposed in [33] was a model for trust management of the internet of vehicles. It was based on context-based cognition and a Bayesian machine learning model.
The presented method in [34] was an integrated algorithm for misbehavior detection. It was a combination of CNN and LSTM models to reconstruct the location information and an SVM as a binary classification method to check to compromise a vehicle. They used the VeReMi dataset to evaluate the performance of their algorithm.
A stable cluster head selection was presented in [35] using DL spectrum sensing. In the research, a classification problem was designed using Long Short-Term Memory (LSTM) for spectrum sensing.
Anomaly detection of VANETs communications and classification of abnormal behaviors were performed in [36] using DL neural network and LSTM model. They used the VeReMi [37] dataset to detect abnormal behaviors on roads and classify the types of location anomalies.
In [38], a machine learning model was proposed to improve the IDSs performance for preventing attacks in VANET using RF and a posterior detection based on coresets. Their goal was improving the detection accuracy and increasing the detection efficiency.
An RL model was proposed in [39] for routing to have a more stable communication system and increase communication reliability in VANETs. A complex objective space with geo-positioning information of vehicles, propagation signal strength, and environmental path loss with obstacles was created to train the model and get the best route based on route stability and hop count.
The proposed system in [40] was an effective IDS using Adaptive Neuro-Fuzzy Inference System (ANFIS) and CNN. The system was an intelligent IDS using soft computing techniques to detect both known and unknown attacks.
Attack classification in VANETs was performed in [41] using a hybrid optimization-based Deep Maxout Network (DMN). The DMN was used for attack classification, and a hybrid optimization algorithm was designed to select the cluster head, perform routing, and teach the DMN.
The novel proposed method in [42] predicted node mobility in ad hoc networks. An RNN-based method comprising a single feed-forward layer was used for more precise and realistic mobility predictions by depicting interactions and connections between nodes.
The proposed approach in [43] was a secure and intelligent message-forwarding strategy based on DRL. The key supporting elements of the model were reasonably designed, and the model's sufficient training was carried out by DQNs.
Table 1 summarizes the applications of machine learning and DL in VANETs. Based on the table DL model was not used previously for the reliability of vehicular networks, which is used in this paper for CR-VANETs' reliability prediction.
| Reference | Algorithm | Application |
|---|---|---|
| Lv (2015) [15] | DL | Traffic flow prediction |
| Kang (2016) [16] | DNN | Intrusion detection |
| Marzak (2016) [17] | ANN method | Clustering |
| Atallah (2017) [18] | DQN | Safety and QoS |
| Ghaleb (2017) [19] | ANN method | Misbehavior detection |
| Dairi (2018) [20] | DSA | Obstacle detection |
| Hoel (2018) [21] | DQN | Lane change decisions |
| Ye (2018) [22] | RL | Resource allocation |
| Jindal (2018) [23] | CNN | Traffic control and routing |
| Bagherlou (2018) [24] | SA and NN | Clustering |
| Liu (2019) [25] | Naïve bayes classifier | Classification and clustering |
| Karabulut (2019) [26] | ANN | Finding the optimum size of CW |
| Adhikary (2020) [27] | SVM | DDoS attack detection |
| Abdellah (2020) [28] | LSTM | Traffic prediction |
| Li (2020) [29] | RNN | Intrusion detection |
| Vitalkar (2020) [30] | DBN | Intrusion detection |
| Bangui (2021) [31] | RF | Intrusion detection |
| Husnain (2021) [32] | Whale's nature-inspired | Clustering |
| Rehman (2021) [33] | Bayesian learning | Trust management |
| Hsu (2021) [34] | CNN, LSTM, SVM | Misbehavior detection |
| Kareem (2022) [35] | LSTM | Spectrum sensing |
| Xiangyu (2022) [36] | LSTM | Anomaly detection |
| Bangui (2022) [38] | RF | Intrusion detection |
| Teixeira (2022) [39] | RL | Routing |
| Karthiga (2022) [40] | ANFIS and CNN | Intrusion detection |
| Kaur (2022) [41] | DMN | Attack detection |
| Yeruva (2023) [42] | RNN | Mobility forecasting |
| Liu (2023) [43] | DRN and DQN | Secure message forwarding |
| Proposed Method | DL | CR-VANETs' reliability estimation |
ANNs are mathematical tools that imitate biological neural networks like the human brain. These networks provide a nonlinear mapping between inputs and outputs without any detailed knowledge [44]. Generally, a neural network acts as a black box and learns to predict the values of specific output variables given sufficient input.
The DL is a subfield of ANNs introduced by a deep auto-encoder [45]. A DNN is an ANN with multiple layers between the input and output layers [46], enabling it to derive high-level functions from input information. In the DNN architecture, each layer receives input from the previous layer and provides output to the next layer. Current advancements in DL have led to the development of increasingly complex neural network structures, thus offering more powerful learning abilities.
A DL model is proposed in this paper to estimate CR-VANET's reliability. The steps taken when designing the proposed model (Fig. 1) are as follows. First, the required data is identified. Then, the model structure (including the inputs, the number and type of the hidden layers, and the model's output) is determined. After that, the DL model parameters (including the number of hidden layers, the number of neurons, activation function, learning rate, batch size, and the number of epochs) are determined. Finally, the DL model is trained and validated.
Dataset generation
One of the most critical issues and active research subjects in machine learning is data gathering [47]. Currently, there is no existing dataset for predicting the reliability of vehicular networks. Thus, a dataset is provided based on a previous method for CR-VANETs' reliability estimation. Most previously proposed methods do not consider driver behavior when estimating VANETs' reliability despite it being one of the most important safety issues in the ITS. Hence, a dataset is generated based on the proposed analytical framework in [14], which estimates the reliability of CR-VANETs with drivers in a highway environment. The researchers who proposed this framework define reliability as the reliable and timely transmission and reception of messages and accurate and on-time reactions of the driver as they navigate the vehicle. Hence, the driver is one of the vehicular nodes' components considered for estimating reliability using the proposed framework.
The Primary Reliability Factors (PRFs) provide a comprehensive basis for characterizing the conditions under which failures occur. The PRFs conditions are considered inputs, and the general reliability of the network is the output in the generated dataset (Fig. 2) for our proposed DL method based on the application of the method in [14] for a CR-VANET with drivers. A dataset is constructed with 1500 rows of data, including different levels for the nine inputs and the estimated total reliability of the CR-VANET with drivers.
The DL model's structure and parameters
After many experiments, we found that the best architecture for the DL model in this paper is fully connected using LSTM. LSTM is an RNN architecture [48] used in DL. A standard LSTM neuron consists of an input gate (it), a forget gate (ft), a cell (ct), an output gate (ot), and an output response (ht) [49] (Fig. 3). The cell remembers values over time intervals, and the gates regulate the flow of information into and out of the cell.
The proposed DL model has nine inputs, two LSTM hidden layers, and an output layer that gives the reliability predictions (Fig. 4). The inputs consist of the PRFs' conditions that affect the CR-VANET's reliability (see Fig. 2).
Considering numerous factors determining different parameters of the DL model is essential; however, no sufficient guidelines are available for selecting parameters. Hence, the parameters in this paper are determined by trial and error to attain acceptable results. An activation function's purpose is to convert an input signal to an output signal of a node. Here, Rectified Linear Units (ReLU) is used as the activation function (Fig. 5). The utilized learning rate for adjusting the model's weights and beta_1 are set at 0.001 and 0.9, respectively. After several experiments, two hidden layers are chosen because adding more layers did not make better results than those obtained using two layers. Moreover, the number of neurons, batch size, and epochs are set at 180, 100, and 10, respectively. Table 2 summarizes the used parameters for the proposed DL model.
|
Parameter |
Value |
|---|---|
|
Activation function |
ReLU |
|
Learning rate |
0.001 |
|
beta_1 |
0.9 |
|
Number of hidden layers |
2 |
|
Number of neurons |
180 |
|
Batch size |
100 |
|
Number of epochs |
10 |
Model evaluation indexes
There are many statistical indexes for evaluating the performance of machine learning methods. The three most often used indexes in previous research are used in this paper, calculated by Equations (1)–(3) [50], [51]:
-
The Correlation Coefficient (R) determines the relationship between the two variables. The closer the index value is to 1, the stronger the direct correlation relationship between the considered variables.
-
RMSE is a useful index for measuring differences between the values predicted by a model and the observed values
-
MAE is an arithmetic average of absolute errors.
where \({x}_{i}^{o}\) and \({x}_{i}^{p}\)are observed and predicted parameters, and nd is the number of data.
This section provides the results of the proposed DL method and compares them with the other machine learning methods (i.e., linear regression, random forest, support vector machine with three kernel functions, and bagging) applied to the generated dataset. In the utilized dataset, 70% of the data is used for training and the remaining 30% for testing. WEKA3.9.0 [52] is used to generate the considered models to estimate the reliability of the CR-VANET with drivers.
Results of the proposed DL model
The evaluation results of the proposed DL model are presented in Table 3. The results show that the DL model effectively estimates the considered CR-VANET's reliability as evidenced by the high correlation coefficient (more than 98%) and the low RMSE (less than 4%) and MAE (less than 3%) for the test data.
Figures 6 and 7 provide the predicted and measured values diagrams for the training and testing data in the proposed DL method. These figures also show the effectiveness of the proposed model considering the reliability prediction of the considered CR-VANET.
| Parameter | Train | Test |
|---|---|---|
| R | 0.9939 | 0.9862 |
| RMSE | 0.0251 | 0.0381 |
| MAE | 0.0158 | 0.0202 |
Comparison of the proposed DL model with other ML methods
The comparison of the previous machine learning models with our proposed DL model for training and testing data is summarized in Table 4. The results show the superiority of the proposed method over the other machine learning methods. Its superiority is indicated by its higher correlation coefficient and lower RMSE and MAE than the other methods. Moreover, the comparisons for the test data are shown in Figs. 8–10. The following insights are inferred from these figures:
-
The DL model has the highest correlation coefficient, followed by the RF, SVM (RBF), bagging, SVM (PUK), LR, and SVM (polynomial) methods (Fig. 8).
-
The DL model has the lowest level of RMSE, followed by the RF, SVM (RBF), bagging, LR, SVM (polynomial), and SVM (PUK) methods (Fig. 9).
-
The DL model has the lowest level of MAE, followed by the RF, SVM (RBF), bagging, SVM (polynomial), LR, and SVM (PUK) methods (Fig. 10).
All figures show that the DL model is the best-performing method among the examined machine learning methods in terms of the considered evaluation parameters.
| Method | Train | Test | ||||
|---|---|---|---|---|---|---|
| R | RMSE | MAE | R | RMSE | MAE | |
| Linear Regression | 0.9467 | 0.0712 | 0.0547 | 0.938 | 0.0795 | 0.0605 |
| SVM (Polynomial) | 0.9432 | 0.0751 | 0.0521 | 0.9357 | 0.0836 | 0.0603 |
| SVM (PUK) | 1 | 0.0013 | 0.0009 | 0.9463 | 0.0942 | 0.0728 |
| SVM (RBF) | 0.9852 | 0.038 | 0.0218 | 0.971 | 0.0547 | 0.0351 |
| Random Forest | 0.9972 | 0.0174 | 0.0108 | 0.976 | 0.0511 | 0.0309 |
| Bagging | 0.9856 | 0.0374 | 0.0234 | 0.9677 | 0.058 | 0.0351 |
| Proposed DL model | 0.9939 | 0.0251 | 0.0158 | 0.9862 | 0.0381 | 0.0202 |
Reliability estimations of vehicular networks represent a critical issue, and machine learning methods have not yet been used to estimate the reliability of VANETs. Hence, a previously proposed method is used to generate a dataset for the reliability of a CR-VANET while considering drivers as one of the vehicles' components. The dataset comprises 1500 rows of data, including nine inputs and one output that 70% of the data is used for training, and the other 30% for testing. Then, the system's reliability is automatically estimated using the proposed DL model and compared with several machine learning methods, namely LR, SVMs with three kernels, RF, and bagging. The proposed DL method is validated based on its correlation coefficient, RMSE, and MAE. The results indicate that the proposed DL method estimates CR-VANETs' reliability more accurately than the other methods. A suggestion for the future is to provide more datasets and apply the proposed DL model to the new datasets to compare the results.
Ethical Approval
We approve that this paper is original and not submitted to another journal simultaneously.
Competing Interests
We declare that there are no conflicts of interest for this research.
Authors' Contributions
Somayeh Bahramnejad and Naser Movahedinia wrote and reviewed the manuscript. Amineh Naseri performed the experiments and provided the tables and diagrams in experimental results section.
Funding
We declare that no financial contributions or funding were received for the production of this research.
Availability of Data and Materials
The datasets generated during the current study are available from the corresponding author on reasonable request.
- Lin S, Wang Y, Jia L (2018) System reliability assessment based on failure propagation processes, Complexity, pp. 1-19. doi: 10.1155/2018/9502953.
- Kim Y, Kang W (2013) Network reliability analysis of complex systems using a non- simulation-based method, Reliab Eng Syst Safe 110, pp. 80-88. doi: 10.1016/j.ress.2012.09.012.
- Singh C, Jirutitijaroen P, Mitra J (2019) Electric Power Grid Reliability Evaluation: Models and Methods. Wiley-IEEE, pp. 117-164.
- Saajid H, Di W, Memon S. Bux N K (2019) ST reliability and connectivity of VANETs for different mobility environments, KSII T Internet Inf 13 (5), pp. 2338-2356. doi: 10.3837/tiis.2019.05.006.
- Zeadally S, Guerrero J, Contreras J (2020) A tutorial survey on vehicle-to-vehicle communications, Telecommun Syst 73, pp. 469-489. doi: 10.1007/s11235-019- 00639-8.
- Sadatpour V, Zargari F, Ghanbari M (2019) A new cost function for improving anypath routing performance of VANETs in highways, Wirel Netw 25 (4), pp. 1657- 1667. doi: 10.1007/s11276-017-1620-0.
- Satheshkumar K, Mangai S (2020) EE-FMDRP: energy efficient-fast message distribution routing protocol for vehicular ad-hoc networks, Ambient Intelligence and Humanized Computing 12, pp. 3877-3888. doi: 10.1007/s11276-017-1620-0.
- Dua A, Kumar N, Bawa S (2017) ReIDD: reliability-aware intelligent data dissemination protocol for broadcast storm problem in vehicular ad hoc networks, Telecommun Syst 64, pp. 439-458. doi: 10.1007/s11235-016-0184-0.
- Lim J, Naito K, Yun J, Cabric D, Gerla M (2015) Safety message dissemination in NLOS environments of intersection using TV white space, International Conference on Computing, Networking and Communications, pp. 451-455. doi: 10.1109/ICCNC.2015.7069386.
- Huang X, Wu J, Li W, Zhang Z, Zhu F, Wu M (2016) Historical spectrum sensing data mining for cognitive radio enabled vehicular ad-hoc networks, IEEE Transactions on Dependable and Secure Computing 13 (1), pp. 59-70. doi: 10.1109/TDSC.2015.2453967.
- Bkassiny M, Li Y, Jayaweera S K (2013) A survey on machine-learning techniques in cognitive radios, IEEE Commun Surv Tut 15 (3), pp. 1136-1159. doi: 10.1109/SURV.2012.100412.00017.
- Singh K D, Rawat P, Bonnin J M (2014) Cognitive radio for vehicular ad-hoc networks (CR-VANETs): approaches and challenges, EURASIP J Wirel Comm 2014 (49), pp. 1-22. doi: 10.1186/1687-1499-2014-49.
- Gillani M, Niaz H A, Tayyab M (2021) Role of machine learning in WSN and VANETs. Journal of Electrical and Computer Engineering Research 1 (1), pp. 15-20. doi: 10.53375/ijecer.2021.24.
- Bahramnejad S, Movahhedinia N (2021) A fuzzy arithmetic-based analytical reliability assessment framework (FAARAF): case study, cognitive radio vehicular networks with drivers, Computing. doi: 10.1007/s00607-021-00980-4.
- Lv Y, Duan Y, Kang W, Li Z, Wang F Y (2015) Traffic flow prediction with big data: A deep learning approach. IEEE T Intell Transp 16 (2), pp. 865-873. doi: 10.1109/TITS.2014.2345663.
- Kang M J, Kang J W (2016) A novel intrusion detection method using deep neural network for in-vehicle network security, IEEE Vehicular Technology Conference (VTC Spring), pp. 1-5. doi: 10.1109/VTCSpring.2016.7504089.
- Marzak B, El Guemmat K, Benlahmar E, Talea M (2016) Clustering in vehicular ad-hoc network using artificial neural network, International Review on Computers and Software 11 (6), pp. 548-556. doi: 10.15866/irecos.v11i6.9328.
- Atallah R, Assi C, Khabbaz M (2017) Deep reinforcement learning-based scheduling for roadside communication networks, International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks, pp. 1-8. doi: 10.23919/WIOPT.2017.7959912.
- Ghaleb F A, Zainal A, Rassam M A, Mohammed F (2017) An effective misbehavior detection model using artificial neural network for vehicular ad hoc network applications, IEEE Conference on Application, Information and Network Security (AINS), pp. 13-18. doi: 10.1109/AINS.2017.8270417.
- Dairi A, Harrou F, Sun Y, Senouci M (2018) Obstacle detection for intelligent transportation systems using deep stacked autoencoder and k-nearest neighbor scheme, IEEE Sensor 18 (12), pp. 5122-5132. doi: 10.1109/JSEN.2018.2831082.
- Hoel C J, Wolff K, Laine L (2018) Automated speed and lane change decision making using deep reinforcement learning, IEEE International Conference on Intelligent Transportation Systems, 2018, pp. 2148-2155. doi: 10.1109/ITSC.2018.8569568.
- Ye H, Li G Y, Juang B H (2018) Deep reinforcement learning for resource allocation in V2V communications, IEEE International Conference on Communications, pp. 1- 5. doi: 10.1109/ICC.2018.8422586.
- Jindal A, Aujla G S, Kumar N, Chaudhary R, Obaidat M S, You I (2018) SeDaTiVe: SDN-enabled deep learning architecture for network traffic control in vehicular cyber- physical systems, IEEE Network 32 (6), pp. 66-73. doi: 10.1109/MNET.2018.1800101.
- Bagherlou H, Ghaffari A (2018) A routing protocol for vehicular ad hoc networks using simulated annealing algorithm and neural networks, Journal of Supercomputing 74, pp. 2528-2552. doi: 10.1007/s11227-018-2283-z.
- Liu T, Shi S, Gu X (2019) Naive bayes classifier based driving habit prediction scheme for VANET stable clustering, in: Artificial intelligence for communications and networks, S. Han, L. Ye, and W. Meng, eds., AICON 2019. Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, 286. Springer, Cham., 2019. doi: 10.1007/978-3-030-22968-9_40.
- Karabulut M A, Shahen S A F M, Ilhan H (2019) Performance optimization by using artificial neural network algorithms in VANETs, International Conference on Telecommunications and Signal Processing (TSP), pp. 633-636. doi: 10.1109/TSP.2019.8768830.
- Adhikary K, Bhushan S, Kumar S, Dutta K (2020) Hybrid algorithm to detect DDoS attacks in VANETs, Wireless Pers Commun 114, pp. 3613-3634. doi: 10.1007/s11277- 020-07549-y.
- Abdellah A R, Koucheryavy A (2020) VANET Traffic Prediction Using LSTM with Deep Neural Network Learning, O. Galinina et al. (Eds.): NEW2AN 2020/ruSMART 2020, LNCS 12525, pp. 281–294.
- Li F, Zhang J, Szczerbicki E, Song J, Li R, Diao R (2020) Deep learning-based intrusion system for vehicular ad hoc networks, Computers, Materials & Continua 65 (1), pp. 653-681. doi:10.32604/cmc.2020.011264.
- Vitalkar R S, Thorat S S, Rojatkar D V (2020) Intrusion detection system for vehicular ad-hoc network using deep learning, International Research Journal of Engineering and Technology 7 (12), pp. 2294-2300.
- Bangui H, Ge M, Buhnova B (2021) A hybrid data-driven model for intrusion detection in VANET, Procedia Comput Sci 184, pp. 516-523. doi: 10.1016/j.procs.2021.03.065.
- Husnain G, Anwar S (2021) An intelligent cluster optimization algorithm based on whale optimization algorithm for VANETs (WOACNET), PLoS ONE 16(4), e0250271. doi: 10.1371/journal.pone.0250271.
- Rehman A, Hassan M F, Hooi Y K, Qureshi M A, Chung T D, Akbar R, Safdar S (2021) Context and machine learning based trust management framework for Internet of vehicles, Computers, Materials & Continua 68 (3), pp. 4125-4142. doi:10.32604/CMC.2021.017620.
- Hsu H Y, Cheng N H, Tsai C W (2021) A Deep Learning-Based Integrated Algorithm for Misbehavior Detection System in VANETs. In ACM International Conference on Intelligent Computing and its Emerging Applications (ACM ICEA ’21), December 28– 29, Jinan, China. ACM, New York, NY, USA, 6 pages.
- Kareem J M, Trabelsi H (2022) A Novelty of Hypergraph Clustering Model (HGCM) for Urban Scenario in VANET, IEEE Access, Vol. 10, No. 2022, pp. 66672-66693. doi: 10.1109/ACCESS.2022.3185075.
- Xiangyu L (2022) Misbehavior Detection based on Deep Learning for VANETs, International Conference on Networks, Communications and Information Technology (CNCIT), pp. 122-128.
- Heijden R W, Lukaseder T, Kargl F (2018) VeReMi: A Dataset for Comparable Evaluation of Misbehavior Detection in VANETs. In: Beyah R, Chang B, Li Y, Zhu S (eds) Security and Privacy in Communication Networks. Secure Comm 2018. Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, Vol. 254. Springer, Cham. doi: 10.1007/978-3-030-01701-9_18.
- Bangui H, Ge M, Buhnova B (2022) A hybrid machine learning model for intrusion detection in VANET. Computing 104, pp. 503–531. doi: 10.1007/s00607-021-01001-0.
- Teixeira L H, Huszák Á (2022) Reinforcement Learning Environment for Advanced Vehicular Ad Hoc Networks Communication Systems. Sensors 2022, 22, 4732. doi: 10.3390/s22134732.
- Karthiga B, Durairaj D, Nawaz N, Venkatasamy T K, Ramasamy G, Hariharasudan A (2022) Intelligent Intrusion Detection System for VANET Using Machine Learning and Deep Learning Approaches, Wirel Commun Mob Com, pp. 1-13.
- Kaur G, Kakkar D (2022) Hybrid optimization enabled trust-based secure routing with deep learning-based attack detection in VANET, Ad Hoc Netw, 136 (2022). doi: 10.1016/j.adhoc.2022.102961.
- Yeruva A R, Alomari E S, Rashmi S, Shrivastava A, Kathiravan M, Chaturvedi A (2023) A Secure Machine Learning-Based Optimal Routing in AdHoc Networks for Classifying and Predicting Vulnerabilities, Cybernet Syst. doi: 10.1080/01969722.2023.2166241.
- Liu B, Xu G, Xu G, Wang C, Zuo P (2023) Deep Reinforcement Learning-Based Intelligent Security Forwarding Strategy for VANET. Sensors 2023, 23, 1204. doi: 10.3390/s23031204.
- Medina E A, Paredes J P (2009) Artificial neural network modeling techniques applied to the hydro desulfurization process, Math Comput Model 49 (1), pp. 207-214. doi: 10.1016/j.mcm.2008.05.010.
- Tong W, Li L, Zhou X, Hamilton A, Zhang K (2019) Deep learning PM2.5 concentrations with bidirectional LSTM RNN, Air Qual Atmos Hlth 12 (4), pp. 411- 423. doi: 10.1007/s11869-018-0647-4.
- Deng L, Yu D (2014) Deep learning: methods and applications, Foundations and Trends in Signal Processing 7 (3-4), pp. 1-199. doi: 10.1561/2000000039.
- Roh Y, Heo G, Whang S E (2021) A survey on data collection for machine learning: a big data-AI integration perspective, IEEE T Knowl Data En 33 (4), pp. 1328-1347. doi: 10.1109/TKDE.2019.2946162.
- Hochreiter S, Schmidhuber J (1997) Long short-term memory, Neural Comput, 9 (8), pp. 1735-1780. doi: 10.1162/neco.1997.9.8.1735.
- Guillén-Navarro M A, Martínez-España R, Llanes A, Bueno-Crespo A, Cecilia J M (2020) A deep learning model to predict lower temperatures in agriculture, Journal of Amb Int Sm Env 12 (1), pp. 21-34. doi: 10.3233/AIS- 200546.
- Tohidi S, Sharifi Y (2016) Load-carrying capacity of locally corroded steel plate girder ends using artificial neural network, Thin Wall Struct 100, pp. 48-61. doi: 10.1016/j.tws.2015.12.007.
- Boukerche A, Wang J (2020) A performance modeling and analysis of a novel vehicular traffic flow prediction system using a hybrid machine learning-based model, Ad Hoc Netw 106, pp. 1-10. doi: 10.1016/j.adhoc.2020.102224.
- Witten I H, Frank E, Trigg L, Hall M, Holmes G, Cunningham S J (1999) Weka: practical machine learning tools and techniques with Java implementations, Workshop on Emerging Knowledge Engineering and Connectionist-Based Information Systems, pp. 192-196.
No competing interests reported.