Peer-reviewed framework combines ML, fuzzy logic and reinforcement learning for self-healing sensor networks

Authors: Manoj Kumar Reddy Devarla and S. V. N. Santhosh Kumar, Vellore Institute of Technology · Publication: 26 July 2026

CWSN-AI v4.2 combines LightGBM fault classification, a Mamdani fuzzy inference system, Dyna-Q routing and trust-based intrusion detection. The authors report 97.4% classification accuracy on a synthetic 200,000-sample dataset, a 99.2% average attack-detection rate, improved packet delivery and longer network life than five comparison protocols.

Why it matters: It illustrates a modular, closed-loop architecture for constrained edge networks. Metrics are author-reported and largely simulation/synthetic-data based; field validation is required.

Paper: Scientific Reports