Efficient Cyber Threat Detection in Smart Home IoT Networks Using Machine Learning and Explainable AI

Abstract

Smart homes are increasingly common, making it more important than ever to safeguard them against cyber threats. In this work, we develop an improved anomaly-based Intrusion Detection System (IDS) for smart home IoT networks by combining machine learning (ML) techniques with advanced Feature Selection (FS) and Explainable Artificial Intelligence (XAI). We propose an FS approach that finds the intersection of important features identified by Recursive Feature Elimination with Cross-Validation (RFECV) using two different base learners: a Decision Tree (DT) and a Light Gradient Boosting Machine (LGBM). This yields a compact subset of 12 network traffic features. Using this significantly reduced feature set, our lightweight LGBM model achieves 99.86% detection accuracy, an F1-score of 99.93%, and a recall of 99.96% on the IoTID20 dataset. The false positive rate is greatly reduced, and computational cost is also minimized. To enhance model transparency, we integrate XAI tools, Local Interpretable Model-Agnostic Explanations (LIME) provide clear instance-level explanations, and Shapley Additive Explanations (SHAP), which provide global explanations of the classifier’s decisions. Experimental results demonstrate that combining FS with XAI can improve both the efficacy and the usability of IoT IDS. Our hybrid model outperforms several recent deep learning approaches in precision and recall, while being far more interpretable. Overall, the proposed method yields high detection rates with improved transparency, making it well-suited for resource-constrained smart home environments.