Fault diagnosis based on extremely randomized trees in wireless sensor networks

Abstract

Abstract Wireless Sensor Network (WSN) being highly diversified cyber–physical system makes it vulnerable to numerous failures, which can cause devastation towards safety, economy, and systems’ reliability. Precise detection and diagnosis of failures or faults in WSN is a challenging issue due to the diversity of deployment and the limitations in the sensors’ resources. In this paper, supervised machine learning-based technique is considered to scrutinize the behavior of sensors through their data for the detection and diagnosis of faults. Most of the faults that commonly occur in WSN are considered: hardover, drift, spike, erratic, data-loss, stuck, and random fault. A trusted dataset published online by the researchers at the University of North Carolina composed of temperature and humidity sensor measurements of multi-hop scenario was acquired and the aforementioned faults were simulated in non-faulty (normal) data. Events from fault occurrences were generated to replicate realistic scenarios of WSN. To detect and diagnose the faults in timely manner, we adopt an ensemble learning-based lightweight technique called Extremely Randomized Trees or Extra-Trees. The proposed Extra-Trees-based detection scheme has the ability of robustness towards signal noise and strong reduction of bias and variance error. The performances of the proposed scheme were compared with those of the state-of-the-art machine learning algorithms such as support vector machine, random forest, neural network, and decision tree. Performance evaluation shows the efficiency of the proposed scheme in terms of accuracy, precision, and F1-score. In addition, the proposed scheme has low training time compared to state-of-the-art approaches.