SHAP based hybrid intrusion detection scheme using ADASYN, filter method, and ensemble learning for IoT and computer network

Abstract With the fast growth of digital infrastructures and the extensive adoption of Internet of Things (IoT) devices, modern networks are increasingly exposed to a diverse range of cyber threats that challenge their security and reliability. An Intrusion Detection System (IDS) acts as a vigilant guardian, monitoring network activity to uncover hidden threats and safeguard the digital environment through timely and effective responses. AI-driven sophisticated models may detect new and unknown threats but, these complex models are mostly like black-box and have limitations with respect to transparency and interpretability in the decision-making process. To address the current scenario, there is a need to develop a hybrid model that amalgamates the interaction between human and technology to create a sustainable and resilient system in Industry 5.0. Here, an explainable AI-based hybrid IDS framework with a feedback mechanism is proposed. To address the data imbalance issue, an attempt has been made using a data sampling technique. To ensure scalability and rapid cost-effective computation, feature selection being performed. Issues of bias-variance and the single learner problem, an ensemble model has been suggested. For the decision process, SHapley Additive exPlanations(SHAP) being attributed to comprehend feature’s involvement. Experiments imparted based on benchmark datasets - CICIDS2017 and MQTT-IoT-IDS2020. Different parameter configurations employed during the experiment to evaluate the behaviour and outcomes under diverse conditions. Further, a comparative analysis demonstrates that this proposed innovative model excels the present models in various performance metrics.