Causal Meta-Learning for Explainable Zero-Day Attack Attribution in Software-Defined Industrial Networks
S. Saranya, S. Rosy Christy, G. Simi Margarat, V Sheeja Kumari, S. Brintha Rajakumari · 2025
A causal meta-learning framework serves as the main approach in this study for explaining zero-day attack attribution in software-defined industrial networks (SDINs) by correcting the weaknesses of correlation-based intrusion detection systems. The proposed method joins meta-learning capabilities for promptly detecting new threats with causal inference capabilities for generating understandable attack tree visualizations which allows precise and trackable attack hazard detection across industrial control system environments. The research introduces three essential enhancements which consist of (1) a causal graph prototype network for attack pattern learning from sparse data and (2) automatic security system adaptation to SDIN threats including power grids and 5G-OT environments and (3) operational validation achieving superior attribution efficiency with 15% enhancement versus LSTM/GNN and forty percent faster adaptation relative to MAML/Proto Net meta-learning platforms. The offered framework decreases wrong attribution identifications by 22% besides delivering evidence human beings can understand which addresses the gap between adaptive machine learning techniques and explainable AI (XAI) for critical infrastructure cybersecurity protection. The study utilizes Causal meta-learning to analyse zero-day attacks through attack attribution of software-defined industrial networks using explainable AI (XAI) in adaptive cybersecurity.