Machine Learning for Fault Detection in Wireless Sensor Networks: A Survey

Machine learning (ML) has become central to modern fault detection in Wireless Sensor Networks (WSNs), enabling improved robustness over traditional threshold-based schemes. However, deploying ML in WSNs remains a challenging task due to the tight constraints of sensor nodes, the absence of standardized datasets, and the growing need of interpretable and energy-efficient decision-making in safety-critical environments. This survey provides a structured and up-to-date review of the state of the art in ML-based fault detection for WSNs, analyzing fault taxonomies, learning paradigms, model placement strategies, dataset limitations, and evaluation practices. Particular emphasis is placed on two emerging challenges: energy efficiency, which affects the feasibility of on-node and cluster-level inference, and interpretability, which is increasingly required for transparent and trustworthy deployments. We compare lightweight and complex ML models, review intrinsic and post-hoc explanation methods, and provide a unified benchmarking perspective through dataset categorization and evaluation criteria. Finally, we highlight persistent gaps and outline a research agenda toward reliable, interpretable, and resource-aware ML frameworks for next-generation WSN fault detection.