<p>Vehicular ad hoc networks (VANETs) represent a crucial element of intelligent transportation systems, enabling real-time communication between vehicles and road infrastructure. These networks enhance traffic management, bolster road safety, and promote the advancement of autonomous driving technologies. Nevertheless, their reliance on wireless communications renders them susceptible to a range of cybersecurity threats, highlighting the necessity for sophisticated intrusion detection solutions (IDS). Traditional intrusion detection systems often rely on machine learning techniques, suffer from low detection accuracy and require extensive feature engineering, limiting their effectiveness in real-time applications. Deep learning-based IDSs have gained popularity due to their ability to automatically extract features from network traffic. However, their performance heavily depends on the effectiveness of the feature learning process. To overcome these issues, we propose, GA-Net, a novel Convolutional Neural Network (CNN) that incorporates a gated mechanism. The gated mechanism in GA-Net plays a crucial role by modulating the flow of information, enabling the model to focus on important features and filter out irrelevant or noisy data. This selective gating improves the model’s ability to capture temporal dependencies, enhancing feature extraction and representation for more accurate anomaly detection in VANET. GA-Net is evaluated on the NSL-KDD dataset using five-fold cross-validation for both binary classification (Normal vs. Attack) and multi-class classification (including DoS, Probe, R2L, U2R attacks, and Normal). The model achieves an average accuracy of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(99.74\%\)</EquationSource> </InlineEquation>, an average F1-score of <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(99.6\%\)</EquationSource> </InlineEquation>, an average precision of <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(99.6\%\)</EquationSource> </InlineEquation> and an average recall of <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(99.59\%\)</EquationSource> </InlineEquation> for the binary classification task. For the multi-class classification task, GA-Net achieves an average accuracy of <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(99.57\%\)</EquationSource> </InlineEquation>, an average F1-score of <InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(99.39\%\)</EquationSource> </InlineEquation>, an average precision of <InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(99.6\%\)</EquationSource> </InlineEquation> and an average recall of <InlineEquation ID="IEq8"> <EquationSource Format="TEX">\(99.19\%\)</EquationSource> </InlineEquation>, demonstrating its robustness in detecting a wide range of attacks. These results demonstrate that GA-Net outperforms existing CNN-based IDS models, providing high detection accuracy, outstanding precision and commendable recall and F1-score across both binary and multi-class classification tasks. By effectively capturing both spatial and temporal dependencies in network traffic, GA-Net offers a scalable and efficient method for securing vehicular networks in dynamic environments.</p>

A Novel Deep Learning-based Intrusion Detection Model for Cybersecurity in Intelligent Vehicular Ad Hoc Networks

期刊论文
发表日期：

2026年2月12日

查看全文

Khadija Mouatassim,
Abdelfettah Mabrouk,
Chaimae Ouchicha

摘要

Vehicular ad hoc networks (VANETs) represent a crucial element of intelligent transportation systems, enabling real-time communication between vehicles and road infrastructure. These networks enhance traffic management, bolster road safety, and promote the advancement of autonomous driving technologies. Nevertheless, their reliance on wireless communications renders them susceptible to a range of cybersecurity threats, highlighting the necessity for sophisticated intrusion detection solutions (IDS). Traditional intrusion detection systems often rely on machine learning techniques, suffer from low detection accuracy and require extensive feature engineering, limiting their effectiveness in real-time applications. Deep learning-based IDSs have gained popularity due to their ability to automatically extract features from network traffic. However, their performance heavily depends on the effectiveness of the feature learning process. To overcome these issues, we propose, GA-Net, a novel Convolutional Neural Network (CNN) that incorporates a gated mechanism. The gated mechanism in GA-Net plays a crucial role by modulating the flow of information, enabling the model to focus on important features and filter out irrelevant or noisy data. This selective gating improves the model’s ability to capture temporal dependencies, enhancing feature extraction and representation for more accurate anomaly detection in VANET. GA-Net is evaluated on the NSL-KDD dataset using five-fold cross-validation for both binary classification (Normal vs. Attack) and multi-class classification (including DoS, Probe, R2L, U2R attacks, and Normal). The model achieves an average accuracy of \(99.74\%\) , an average F1-score of \(99.6\%\) , an average precision of \(99.6\%\) and an average recall of \(99.59\%\) for the binary classification task. For the multi-class classification task, GA-Net achieves an average accuracy of \(99.57\%\) , an average F1-score of \(99.39\%\) , an average precision of \(99.6\%\) and an average recall of \(99.19\%\) , demonstrating its robustness in detecting a wide range of attacks. These results demonstrate that GA-Net outperforms existing CNN-based IDS models, providing high detection accuracy, outstanding precision and commendable recall and F1-score across both binary and multi-class classification tasks. By effectively capturing both spatial and temporal dependencies in network traffic, GA-Net offers a scalable and efficient method for securing vehicular networks in dynamic environments.