A Hybrid Approach for Anomaly Detection in Autonomous Vehicles Using Genetic Algorithm and GRU Model

Abstract

Autonomous Vehicle (AV) and Connected Autonomous Vehicle (CAV) technologies for transport systems to become safer, less congested and cleaner is a technological breakthrough. Nevertheless, these are complex systems and the interconnection between them presents a number of challenges, more notably in the form of cyber security and Anomaly Detection (AD) from sensor data. Errors due to sensor malfunctions, cyber-attacks and environmental disturbances can disrupt vehicle performance and safety. In this work, we propose the Hybrid model which combines Genetic Algorithm (GA) for feature selection process and Gated Recurrent Unit (GRU) based models for anomaly classification. The GA is used to do so such that the results would be more accurate of detection. where the GRU simulates the processing of time-series sensor data to detect anomalous behaviors. We evaluate this model on the MMITSS-Multi-Modal Intelligent Transportation Signal Systems dataset which is a large-scale real-world dataset with millions of records, and three different feature channel configurations, one feature channel Mean Squared Error (MSE), Two feature channels (MSE, Mean Absolute Percentage Error (MAPE)), three feature channels (MSE, MAPE, original data). The proposed model achieves better performance in terms of accuracy, precision, recall and specificity as compared to existing techniques, especially for detecting anomalies. This hybrid approach is an excellent and scalable means of detecting online anomalies in CAVs, thus contributing to enhance the safety and dependability of self-driving systems.