Leveraging Behavior-Based Anomaly Detection for Protecting Internet of Things (IoT) Devices

The widespread use of Internet of Things (IoT) devices has led to increased exposure to cyber-attacks due to their limited processing and security capabilities. This paper investigates the use of supervised machine learning models in detecting anomalies in IoT environments using the Edge-IIoTset dataset, including labeled network traffic such as botnet activity, DoS attacks, and unauthorized access. Preprocessing techniques such as one-hot encoding, normalization, and correlation-based feature selection were applied. Models tested include Logistic Regression, Naïve Bayes, Random Forest, AdaBoost, XGBoost, and CatBoost. Tree-based ensemble models achieved the highest accuracy, with XGBoost reaching 98.46%. The findings validate the scalability and efficiency of ensemble models for edge-based real-time deployment and emphasize the importance of lightweight, interpretable intrusion detection systems for resource-constrained IoT devices.