Machine Learning Based Trust Aggregation for IoT Systems
Pasindu Manisha Kuruppuarachchi, Alan McGibney, Susan Rea, Bernd-Ludwig Wenning · 2025
IoT systems consist of multiple heterogeneous sensors, actuators, and control logic. These systems not only collect data from the physical world around us, they play a critical role in supporting decision-making processes. Trust is essential in this context, as decision-makers must rely on the system's ability to perform its assigned tasks reliably. To ensure system-level trust, a trust analyser is implemented to assess the trustworthiness of IoT systems across seven distinct evaluation categories. Various tools and techniques can be applied within these categories, and the accuracy of each tool must be considered when aggregating trust evaluations to produce a representative trust score. To address this, several machine learning based aggregation methods are explored and compared, including Adaptive Neuro-Fuzzy Inference Systems (ANFIS), Artificial Neural Networks (ANN), and the Tsetlin Machine (TM). ANN and TM achieved 98% accuracy in correctly detecting trust attacks, while ANFIS achieved 76% accuracy. In addition, both ANFIS and TM offer interpretability, providing valuable insight into how they detect and flag attacks within the IoT system.