Explainable Ensemble-Based Trust Management for IoT Systems

Maintaining the appropriate level of trust for the specific application and time is essential in the ever-changing world of the Internet of Things (IoT). There are many different classification models and there are no “horses for courses” meaning that we cannot say in advance which classifiers might be successful for a given problem or setup. It is therefore common to use an ensemble of different classification models and let the data drive the way in which they are combined. This research presents a novel ensemble deep learning (DL) method aimed at greatly enhancing the detection accuracy of trust management systems (TMS). Our methodology employs a novel ensemble approach that integrates two distinct architectures of convolutional neural networks (CNNs) using an ensemble stacking procedure. We use the predictions from the CNN models as inputs for a meta-learner, which combines these inputs to produce the final outcome. Regularisation approaches facilitate feature selection by optimising the process to highlight only the most significant behavioral-based features of the data, leading to enhanced detection rates and accuracy. Moreover, we employ model-agnostic methods to enhance the interpretability of the model's decisions by providing explicit and comprehensible explanations for each prediction using CNN. In addition, we demonstrate the efficacy of our model by evaluating it on the transformed UNSW NB 15 dataset, which shows a notable improvement during comparative analysis. This study enhances the field of TMS by integrating model ensemble and model-agnostic methods with the