A Machine Learning Approach for Fault Detection and Reliability Investigation in Wireless Sensor Networks

Abstract

Data loss, reduced network lifespan, and decreased accuracy are common consequences of wireless sensor network (WSN) faults. WSN performance requires fault detection to be both accurate and efficient. This study proposes a hybrid fault detection for WSNs by integrating several machine-learning models to improve anomaly classification. Our method compares the strength of each classifier, including random forest (RF), support vector machine (SVM), k-nearest neighbour (KNN), naïve Bayes (NB), convolutional neural network (CNN), and multilayer perceptron (MLP). This approach is the key novelty because it compares traditional ML and deep learning models with hyperparameter optimization and with better optimized classifier performance measurement for different fault cases like offset, gain, stuck at, and out of bounds faults. To validate our proposed model, we carry out Python-based simulations and analyze the accuracy, precision, recall, and computational efficiency of the proposed model compared to the rest of the classifiers. The results show that KNN, RF, and CNN get 100% accuracy for fault types, with KNN taking the least response time. Given the recognized need for additional optimization of real-world deployment, this work shows the usefulness of multi-ML in selecting the optimal one for creating better fault detection in WSNs.