Enhancing IoT Intrusion Detection with XGBoost-Based Feature Selection and Deep Neural Networks

Abstract

The Internet of Things (IoT) networks face noteworthy vulnerabilities to cyber-attacks, mainly restricting from the widespread integration of interconnected smart devices. To protect these networks, robust Intrusion Detection Systems (IDS) play an essential role. This study endeavors to devise an effective system geared towards identifying and thwarting attacks on IoT networks. Using two comprehensive datasets – BoT-IoT and AWID – containing pertinent network traffic and cyber-attack data, the study formulates an IDS that combines optimized feature selection utilizing XGBoost and deep neural networks to boost attack detection abilities. The methodological approach encompasses the collection and preprocessing of IoT network data, followed by the identification of the most influential features using XGBoost. Subsequent evaluation encompasses various supervised machine learning models such as logistic regression, naïve Bayes, catboost, random forest, alongside a CNN-GRU deep learning model. Impressively, the CNN-GRU model structure shows and revealed detection accuracy beyond 99%, meaningfully outstanding conventional ML models used in our experiments. Comprehensive ablation studies meticulously quantify the contributions of pivotal model components, while robustness and strength evaluations against zero-day attacks further attest to the efficacy and results of the CNN-GRU model. Ultimately, the proposed CNN-GRU model emerges as an efficient and accurate IDS solution, poised to support real-world IoT deployments effectively.