Machine learning-based intrusion and anomaly detection for enhancing security in IoT networks using BoT-IoT dataset

Abstract

The fast development of the Internet of Things (IoT) has presented fresh security issues since the great interconnectedness of gadgets makes them more prone to cyberattacks. The objective of this effort is to build a strong machine learning-based intrusion and anomaly detection system to improve IoT network security. With sophisticated supervised learning methods, the main goal is to precisely and effectively identify and classify harmful traffic. Testing the BoT-IoT dataset after extensive preprocessing including outlier treatment, feature normalising, and class balancing via SMote. Several machine learning algorithms—Random Forest, the Support Vector Machine (SVM), K-Nearest Neighbours (KNN), Decision Tree, as well as XGBoost—were evaluated using standard criteria including accuracy, precision, recall, and F1-score. Visualisations such scatter plots, boxplots, and correlation heatmaps let exploratory data analysis (EDA) reveal attack behaviour and feature distributions. With XGBoost and Random Forest attaining over 99.4% accuracy, greatly exceeding other methods, the findings showed great classification accuracy for all models. These results imply that because of their scalability, versatility, and great detection powers, ensemble-based models are fit for IoT contexts. The study offers in the expanding terrain of IoT a scalable, accurate, and real-time intrusion detection solution.