Trust-based attack detection model for connected cars using a Subjective Logic based framework

Multi-model classifiers, commonly referred to as ensemble or voting classifiers, combine the predictions of multiple base learners to produce a final decision. They are widely used to enhance detection performance in applications such as network intrusion detection systems (NIDS). However, the performance of individual models often varies across classes and operational contexts. Traditional voting schemes, including hard and soft voting, rely on fixed model weights, limiting their ability to adapt to class-specific performance variations and inter-model conflict. To address this limitation, this paper proposes a novel binary ensemble classifier based on Subjective Logic, a probabilistic framework that models belief, uncertainty, and source trustworthiness. The proposed method dynamically adjusts the model weights by incorporating both historical model performance and real-time conflict between predictions. The approach is evaluated in an intrusion detection task for connected vehicles networks. Experimental results demonstrate that the proposed subjective-logic based ensemble significantly surpasses traditional hard and soft voting schemes in terms of detection performance, on par with other state-of-the-art dynamic ensembles.