A HYBRID LIGHTWEIGHT AND EXPLAINABLE FEDERATED LEARNING MODEL FOR REAL-TIME INTRUSION DETECTION IN RESOURCE-CONSTRAINED IOT ENVIRONMENTS

Abstract

In recent years, the increasing adoption of Industrial IoT and smart infrastructure has created new challenges in cybersecurity, particularly concerning data privacy and real-time threat detection. This study proposes a lightweight Federated Learning (FL)-based Intrusion Detection System (IDS) that collaboratively trains a neural network model across distributed clients without requiring centralized data collection. Using the UNSW-NB15 dataset—a comprehensive benchmark for modern network threats—we simulate federated training across multiple clients using a compact feedforward neural network. The model is trained locally on each client and updated globally using the Federated Averaging (FedAvg) algorithm. Our approach preserves data privacy while maintaining high detection accuracy.