Improved Particle Swarm Optimization based Bidirectional-Long Short-Term Memory for Intrusion Detection System in Internet of Vehicle

An intelligent vehicle communication network referred to the internet of vehicles (IoV) has been developed by integrating the vehicle network (VN), the internet of things (IoT) and the vehicle-to-everything (V2X) technologies. However, malicious assaults may easily compromise the vehicle communication network. Protecting the integrity of data exchanged between vehicles in an IoV network demands an intrusion detection system (IDS). It is crucial to understand that the distribution of network traffic data is not uniform, and the presence of various forms of network attacks significantly impacts the functioning of IDS. Data imbalance is a prevalent issue in network traffic data and significantly affects the accuracy of detection. This research addresses this challenge by using the Synthetic Minority Oversampling Technique (SMOTE) to introduce a well-proportioned distribution of classes. However, the wide range of network threats is another important issue that contributes to the inadequate performance of intrusion detection models. The majority of existing Deep Learning (DL)-based IDS primarily focus on binary classification and do not achieve substantial detection of advanced attacks. On the other hand, DL models such as Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Bidirectional-LSTM (B_LSTM), etc. encounter several challenges when it comes to constructing a particular framework by carefully choosing the suitable and optimum values of hyper-parameters. This research implements the Chaotic Particle Swarm Optimization (CPSO) technique to effectively optimize several important hyper-parameters of the B_LSTM model. These hyper-parameters include the learning rate, number of neurons in hidden layers, dropout rate, optimizer and activation functions of the B_LSTM layers. The objective is to improve the model attack detection accuracy performance. Implementation of the proposed B_LSTM + CPSO model using the “Car Hacking: Attack & Defence Challenge-2020” datasets demonstrate outstanding performance in detecting attacks in the IoV environment. The experimental findings have been compared with two other optimization strategies, namely the PSO and Firefly Algorithm (FA). Subsequently the findings are also compared to the performance of standard LSTM, LSTM + PSO, LSTM + FA, LSTM + CPSO and B_LSTM algorithms. The evaluation of the statistical comparisons demonstrates that the proposed B_LSTM + CPSO technique outperformed other investigated approaches in attack detection with an accuracy of about 99.94%. The model optimizes hyper-parameters including a learning rate of 0.001, number of hidden units of 128, dropout rate of 0.1, optimizer as ‘Adam’ and layer activation function as ‘relu’. This research presents a helpful resource for assuring the cyber security of the IoV.