Improving IoT Botnet Attack Detection using Machine Learning: Comparative Analysis of Feature Selection Methods and Classifiers in Intrusion Detection Systems

The rising threat of IoT botnet attacks poses a significant cybersecurity challenge, demanding robust detection mechanisms. This study focuses on optimizing the detection of IoT botnet attacks using Machine Learning (ML) through a comprehensive analysis of feature selection methods and ML classifiers, utilizing the N-BaIoT dataset. Derived from authentic traffic data collected from compromised IoT devices, this dataset addresses the scarcity of publicly available botnet-specific datasets in IoT domains. The research evaluates five distinct feature selection methods—Fisher Score, Mutual Information, Chi-Square, Principal Component Analysis (PCA), and Recursive Feature Extraction (RFE)—in conjunction with four ML classifiers—Logistic Regression, Random Forest, Gradient Boosting, and Support Vector Machine. Performance evaluation encompasses accuracy, precision, and F1 score metrics across various combinations of these methods. The findings emphasize the consistent advancements in accuracy observed with Random Forest and Gradient Boosting, the exceptional precision and F1-scores demonstrated by Fisher Score, and the stability of PCA, when paired with Logistic Regression and Support Vector Machine. Conversely, limitations are observed in the performance of Support Vector Machine across metrics and the inferior performance of the Chi-Square feature selection technique. While, the study focuses on a specific dataset, it establishes a comparative benchmark, providing insights into effective methodologies for detecting IoT botnet attacks. Moreover, these outcomes significantly contribute to advancing IoT security measures, particularly by serving as a reference point for researchers aiming to fortify Intrusion Detection Systems (IDS) against evolving cybersecurity threats in IoT environments.