Enhancing IoT Security: A Novel Approach with Federated Learning and Differential Privacy Integration

The Internet of Things (IoT) has expanded to a diverse network of interconnected electronic components, including processors, sensors, actuators, and software throughout several sectors such as healthcare, agriculture, smart cities, other industries. Despite offering simplified solutions, it introduces significant challenges, specifically data security and privacy. Machine Learning (ML), particularly the Federated Learning (FL) framework has demonstrated a promising approach to handle these challenges, specifically by enabling collaborative model training for Intrusion Detection Systems (IDS). However, FL faces some security and privacy issues, including adversarial attacks, poisoning attacks, and privacy leakages during model updates. Since the encryption, mechanisms poses issues like computational overheads and communication costs. Hence, there is need for exploring of alternative mechanism such as Differential Privacy (DP). In this research, we demonstrate an experimental study aiming exploring of FL with DP to secure IoT environment. This study analyzes the effectiveness of DP in horizontal FL setup under Independent and Identically Distributed (IID) pattern. Results on MNIST dataset show promising outcomes; FL with and without employing DP mechanism achieve an accuracy of 98.92% and 98.2%, respectively. Furthermore, the accuracy rate achieved with complex cybersecurity dataset is 93% and 91% before and after employing the DP mechanism. These findings outlines the efficiency of DP in FL framework for improving security and privacy in IoT environment.