Fault Detection in Wireless Sensor Networks Using Autoencoder Classifier

Abstract

Wireless sensor networks (WSNs) are extensively deployed to gather and process data from monitoring environments. Due to their deployment in harsh and unattended conditions, sensor nodes are highly susceptible to faults, which can have severe consequences on safety, economy, and system reliability. To tackle this challenge, machine learning algorithms have emerged as a promising solution. In this study, we propose a fault detection algorithm based on an autoencoder-based classification model. Unlike traditional neural networks, autoencoders have equal numbers of neurons in the input and output layers, facilitating fault detection by comparing input and output values. We evaluate the proposed algorithm against existing approaches using three key parameters: fault detection accuracy, false alarm rate, and false positive rate. Our simulation results demonstrate the superior performance of the proposed algorithm in detecting faults such as spike, fixed bias, gain, and out-of-bounds faults within the network.