abstract

Wireless Sensor Networks (WSNs) are pivotal in applications ranging from environmental monitoring to industrial automation, yet their reliability is often compromised by node and network faults. This study conducts a comprehensive comparative analysis of machine learning (ML) and deep learning (DL) models for fault diagnosis in WSNs, evaluating performance metrics including accuracy, false positive rate (FPR), computational overhead, and real-time applicability. Utilizing synthetic and real-world datasets, we tested SVM, Random Forest, k-NN, CNN, LSTM, and Autoencoders. Results indicate DL models, particularly CNNs, achieve superior accuracy (95%) but incur higher computational costs, whereas ML models like Random Forest offer a balance between accuracy (89%) and efficiency. This paper provides actionable insights for selecting models based on application constraints, contributing to optimized WSN reliability.