AI-POWERED CYBERSECURITY THREAT PREDICTION AND MITIGATION IN IOT DEVICES: A COMPARATIVE ANALYSIS OF FEDERATED LEARNING MODELS

Abstract

The Internet of Things (IoT) devices across different industries were met with new challenges in the field of cyber security. Therein lies the need for safe and effective threat analysis and detection. This paper focusses on the comparative analysis of different models of cyber security threat prediction and mitigation with the utilization of federated learning in IoT. The study employed three types of models; Federated Learning Averaging (FedAvg), Federated Learning with proximal regularization (FedProx) and a new advanced form of federated learning, and extends over a period of six months, with data traffic from 500 diverse IoT devices in smart homes, factory settings and medical healthcare. The study set up virtual environments in which cyber-attacks were staged, including DDoS, malware injection and attempts at unauthorized access, so as to measure the effectiveness of the models employed. The findings exhibited a remarkable 96.8% accuracy, 95.4% precision and 97.2% recall from the advanced federated learning form, making it the best of the three. The differences in models contributed to the federated learning cyber security frameworks a new equilibrium of security with reasonable computational balance likely to be exceedingly helpful in the low resource settings in IoT. The ANOVA tests statistically validated the differences models used and the performance impact to be considerable. The study culminated to a demonstration of the potential federated learning holds in developing systems for cyber security protection in the distributed detection of IoT ecosystems and systems in location data protection.