Integrating Two-Tier Optimization Algorithm With Convolutional Bi-LSTM Model for Robust Anomaly Detection in Autonomous Vehicles

Industrial development has changed vehicles of traditional into autonomous vehicles (AVs). AVs play a significant role since they are measured as a vital module of smart cities. The AV is an advanced automobile efficient in preserving secure driving by evading collisions formed by drivers. In contrast with traditional vehicles, which are fully coordinated and functioned by humans, AVs gather information regarding the exterior environment utilizing sensors to guarantee secure navigation. AVs decrease environmental effects since they regularly employ electricity to function rather than fossil fuel, thus diminishing greenhouse gasses. However, AVs might be exposed to cyber-attacks, causing dangers to human life. Machine learning (ML) and deep learning (DL) based anomaly recognition has progressed as a new study track in autonomous driving. ML and DL-based anomaly detection scholars have focused on improving accuracy as a typical classification task without aiming at mischievous information. This article develops an improved whale optimization algorithm-based feature selection using explainable artificial intelligence for robust anomaly detection (IWOAFS-XAIAD) technique in autonomous driving. The major aim of the IWOAFS-XAIAD technique is an endwise XAI structure to construe and imagine the anomaly recognition classifications prepared by AI models securing autonomous driving systems. Initially, the IWOAFS-XAIAD technique utilizes the Z-score data normalization method to convert input data into a compatible layout. Besides, the IWOAFS-XAIAD technique employs an improved whale optimization algorithm (IWOA)-based feature subset selection to pick an optimum set of features. An attention mechanism with the CNN-BiLSTM (CNN-BiLSTM-A) model is employed for anomaly detection and classification. Moreover, the catch-fish optimization algorithm (CFOA) selects the hyperparameters connected to the CNN-BiLSTM-A model. Finally, utilizing the SHAP XAI method, the IWOAFS-XAIAD technique performs local and global descriptions for the black-box AI model. To demonstrate the optimum classification outcome of the IWOAFS-XAIAD technique, a wide range of experiments is performed on a VeReMi dataset. The experimental validation of the IWOAFS-XAIAD technique portrayed a superior accuracy value of 98.52% over the existing methods.