A Framework for Evaluating the Effectiveness of Explainability Methods in Deep Learning

The Internet of Things (IoT) is a new framework that is transforming societies across the globe into smart towns. Nevertheless, because of IoT systems' ongoing connection and data exchange, their rise has been followed by several cybersecurity problems. In response to this issue, several intrusion detection structures (IDSs) have been developed as safety precautions. These methods have shown effectiveness in thwarting several IoT-based assaults, particularly when deep learning (DL) techniques are employed. While there is a lot to discuss on the topic of eXplainable Artificial Intelligence (XAI) nowadays, more research is still required to completely comprehend how effective XAI is at identifying assault areas and routes when implemented in cyber safety uses. This research provides an innovative, explainable IDS for IoT systems. This study designed an IDS utilizing a Short-Term Long Memory (LSTM) framework to detect intrusions and analyze the system's choices. For training and assessing the LSTM algorithm, an innovative collection of source characteristics is gathered using a new SPIP (S- Shapley Addition ExPlanations, P- Permuting Characteristic Significance, I- Individual Conditioned Expectations, P- Partial Dependent Graph) architecture. When contrasted with existing competitor approaches, the SPIP architecture outperformed them in terms of identification accuracy, processing duration, and comprehension of data characteristics and models. The suggested method could help managers and decision-makers comprehend complicated assault activity.