Lightweight Federated Supervised Machine Learning for IoT Botnet Detection

The growing number of IoT devices has made them a frequent target for botnet attacks, yet most existing detection methods depend on centralized systems that raise privacy concerns and require high computational resources. This study proposes a lightweight, privacy-preserving botnet detection framework using supervised machine learning within a federated learning setup. We evaluated four models—Decision Tree, K-Nearest Neighbors, Support Vector Machine, and Logistic Regression—on the N-BaIoT dataset across seven IoT nodes. Results show that Decision Trees provide the best trade-off between accuracy and training time, making them suitable for resource-constrained environments. We also demonstrate that combining local models through a simple ensemble improves generalization and achieves up to 98.97% accuracy without sharing sensitive data. This framework offers a practical and scalable solution for securing IoT networks against evolving threats.