Comparative Analysis of Different Attention Mechanisms in Hybrid Deep Learning Model for In Vehicle Intrusion Detection Systems

The increasing connectivity in vehicles makes them more vulnerable to threats. Deep Learning has become a powerful tool for enhancing in-vehicle network security, the attention mechanism plays an important role in improving Intrusion Detection Systems. This study evaluates the performance of Luong, Bahdanau and self-attention mechanisms on a LSTM-GRU based Intrusion detection system. This model is tested using HCL-Car hacking dataset which represent different attack scenarios such as DoS, Fuzzy, Gear, and RPM attacks. The Luong attention mechanism achieved the accuracy of 98.55% and F1 score of 98.70% over 30 epochs with a 64 batch size and bahdanau achieved accuracy of 96.36% and F1 score of 97.10% under similar condition but under similar condition the self-attention mechanism showed low accuracy of 85.01% and F1 score of 89.50%. These results demonstrate how important attention mechanisms are for improving IDS performance for connected vehicles.