Enhancing the Detection of Botnet Attacks in the Internet of Things Networks Through the Utilization of Hybrid Feature Selection

The proliferation of Internet of Things (IoT) technology is leading to an escalation in cyber risks. The Intrusion Detection System (IDS) is crucial in countering cyber threats, with botnets being the most commonly detected attack in IoT networks. Anomaly-based IDS is necessary to enhance their capability in identifying cyber threat attributes through the use of supervised machine learning. This work investigates an enhanced supervised machine learning algorithm technique that combines filter-based feature selection methods, specifically Low Variance and Pearson Correlation Coefficient, to detect a new type of botnet attack. The CICIoT2023 dataset, an updated benchmark intrusion dataset specifically designed for IoT networks, is employed to assess the model's performance using several algorithms including Decision Tree, Random Forest, Logistic Regression, and Naive Bayes. The results demonstrate that the model effectively enhances the identification of botnet attacks by reducing the number of features from 46 to 14. This optimization leads to improved performance, including increased accuracy level and reduced computation time.