Optimizing Botnet Detection in IoT Networks: Feature Selection Analysis on the UNSW-NB15 Dataset

The increasing amount of network traffic data in the context of rapidly expanding IoT networks presents a major challenge to traditional Intrusion Detection Systems (IDS). In order to meet this problem head-on, intelligent and robust intrusion detection systems that can handle false alarms and guarantee better detection accuracy are desperately needed, especially in light of the growing risks posed by botnets. Using the large UNSW-NB15 dataset and machine learning techniques, this study presents a new approach for IoT intrusion detection. In order to defend against common Botnet attacks, our primary objective is to compare several feature selection algorithms, such as SelectKBest, Lasso Regression, Recursive Feature Elimination, and Mutual Information Criterion, with a variety of machine learning models, such as k-Nearest Neighbors (KNN), Random Forest, Decision Tree, and Logistic Regression. By carefully analyzing these feature selection methods, we hope to identify the most effective approach to improve intrusion detection precision in the ever-changing IoT environment. Through focused tests using a carefully selected dataset, this study confirms that feature selection techniques, when used in conjunction with KNN, Ran-dom Forest, Decision Tree, and Logistic Regression, strengthen the capabilities of Internet of Things intrusion detection systems.