Mitigating IoT Botnet Attacks in Smart Homes with Federated Learning-Based Intrusion Detection Systems

Abstract

The proliferation of Internet of Things (IoT) devices in smart homes has created new attack surfaces for malicious actors. Botnets such as Mirai have exploited these vulnerabilities, compromising devices for distributed denial-of-service (DDoS) attacks and other cyber threats. Traditional centralized intrusion detection systems (IDS) face privacy and scalability limitations in smart home environments due to sensitive user data and heterogeneous device types. This paper proposes a federated learning (FL)-based IDS framework for smart homes that allows distributed training of anomaly detection models across IoT devices while preserving data privacy. The framework focuses on detecting botnet-related traffic and anomalous behavior, mitigating threats like Mirai botnets. We implement and evaluate the approach using real-world IoT datasets, including the UNSW-NB15 and Bot-IoT datasets, assessing detection accuracy, false positive rate, and communication overhead. Experimental results demonstrate that the proposed FL-based IDS achieves detection rates above 94% with false positive rates under 3%, outperforming traditional centralized IDS models while maintaining user data privacy.