Federated Learning-Driven Intrusion Detection for Cybersecurity in Smart Distribution system
Peenal Gupta, Banhirup Sengupta, Susham Nandi · 2024
The rapid progress in smart city technologies, facilitated by extensive deployment of Information and Communication Technology (ICT), enhances urban service delivery and resource management. However, this heightened connectivity exposes critical infrastructure to cyberattacks, posing substantial risks to public safety and economic stability. Traditional intrusion detection systems (IDS) often struggle with issues related to data privacy, scalability, and real-time threat detection in decentralized environments. To address these challenges, we propose an innovative intrusion detection system utilizing Federated Learning (FL) combined with Artificial Neural Networks (ANNs) to effectively detect anomalies while preserving data privacy. Our FL-DIDS framework achieves a detection accuracy of 94.7%, indicating a significant improvement over traditional centralized IDS, which typically achieves 80-90% accuracy under controlled conditions. This performance enhancement is achieved through the reduction of false positive rates and detection latency due to localized data processing. Moreover, our system maintains a decreased loss rate from 1.79 to 1.00, reflecting improved prediction accuracy. The consistent improvement in model performance, confirmed by confusion matrix analysis, demonstrates high true positive rates with minimal false positives and negatives, thereby ensuring reliable threat detection in smart distribution systems such as water networks and power grids. This adaptively optimized architecture significantly enhances scalability and robustness within IoT-based smart city infrastructures, providing a flexible real-time intrusion detection and mitigation framework across diverse smart systems.