EX-DFL: An Explainable Deep Federated-based Intrusion Detection System for Industrial IoT
Danish Attique, Hao Wang, Ping Wang, Danish Javeed, Muhammad Adil
Abstract
With the escalating volume of sensitive data trans-mitted in Industrial Internet of Things (1IoT) infrastructures, securing data produced by various IloT devices has become paramount for safeguarding critical industrial processes. In particular, the field of Industrial loT (1IoT) is witnessing an increasing threat landscape, necessitating robust Intrusion Detection Systems (IDS) to establish a secure environment. This paper addresses the security concerns associated with IloT by proposing a Federated Learning (FL)-based IDS. The focus is on preserving data privacy, a critical consideration in industrial settings. The proposed explainable approach employs a Deep Neural Network in FL (EX-DFL) to detect anomalies in IloT traffic, mitigating potential security threats. A key contribution of this work is the incorporation of Explainable AI (XAI) techniques, specifically leveraging the SHapley Additive exPlanations (SHAP) library, to enhance the interpretability of the IDS decisions. This addresses the challenge of understanding the rationale behind the model's predictions in complex industrial environments. The proposed EX-DFL performance is assessed using various standard met-rics and evaluated on the CICIDS2017 dataset, demonstrating promising results by achieving over 99 % accuracy in anomaly detection.