AI Analytics in SDN Network Management: A Review

Abstract

Software-defined networking (SDN) fundamentally reshapes network management by separating control logic from data forwarding, enabling centralized, programmable control at massive scale. Yet this shift brings new challenges including centralized controller vulnerabilities, unpredictable traffic patterns, and quality-of-service demands across diverse infrastructures that traditional rule-based systems struggle to meet. AI and machine learning have emerged as the leading solution, bringing adaptive, data-driven capabilities to SDN management. This systematic review examines AI analytics in SDN management across three high-impact areas: intrusion detection and security, traffic engineering and QoS optimization, and fault detection with self-healing recovery. For each domain, we assess the AI and ML techniques used, their reported performance, integration patterns with SDN architectures, and unresolved challenges. The review finds that hybrid deep learning architectures, notably CNN-LSTM models and transformer-based systems, consistently attain detection accuracies in excess of 99% on established SDN intrusion detection benchmarks. Deep reinforcement learning approaches demonstrate measurable and repeatable advantages over conventional traffic engineering baselines for QoS management. Autonomous self-healing frameworks, while technically promising, remain the least operationally mature domain, constrained by scalability demands and the stringent latency requirements of production environments. Despite impressive lab results, production deployment lags due to poor dataset quality, limited model generalization, high computational costs, and lack of standardized evaluation. We conclude with a research roadmap prioritizing robust datasets, edge deployment strategies, and unified benchmarks to realize AI's potential in operational SDN networks.