Federated learning for sustainable intrusion detection systems: A review of green computing strategies and future directions

Abstract

The rapid proliferation of IoT devices creates dual cybersecurity challenges: traditional centralized IDS consume excessive energy and raise privacy concerns, while federated learning implementations, despite their distributed nature, lack comprehensive sustainability considerations. This study systematically reviews federated learning approaches for intrusion detection systems through a green computing lens, examining how energy efficiency and sustainability can be integrated throughout FL-IDS life-cycle while maintaining robust security. We conducted a systematic literature review of recent FL-IDS implementations and developed a taxonomy that categorizes green computing strategies according to machine learning life-cycle stages: data preparation, local training, aggregation, and inference. Our analysis identified several green computing strategies including model compression techniques, adaptive client selection, energy-aware aggregation protocols, and lightweight inference methods. However, the review reveals that sustainability metrics are inconsistently reported across studies, and carbon footprint assessments remain notably absent from current FL-IDS literature. While federated learning demonstrates potential for sustainable intrusion detection, significant gaps persist between current implementations and fully green cybersecurity systems, highlighting the need for standardized energy metrics, carbon-aware orchestration, and integration with renewable energy sources in future research.