Fault Detection in Wireless Sensor Networks Data Using Random Under Sampling and Extra-Tree Algorithm

As a highly diverse cyber-physical system, Wireless Sensor Network (WSN) is vulnerable to various failures, which can have catastrophic consequences for safety, economy, and system dependability. Due to the various deployments and limitations of sensor resources, proper detection and diagnosis of failures or faults in WSNs is a complex problem. In this study, a supervised machine learning-based approach is used. To address this issue, the authors employ Random Under Sampling (RUS) sampling method, which is used to overcome class imbalance, and Extra-Tree (ET) classification algorithm to examine sensor behavior through data to find and diagnose problems. The performance of the proposed scheme is compared with advanced machine learning algorithms such as Support Vector Machine (SVM) and Random Forest (RF). The efficiency of the suggested scheme is compared based on the measuring parameters of Accuracy, Recall, Precision, F1-Score, and AUC-ROC Score. This study's results showed that the Random Under Sampling (RUS) sampling method could negatively and positively impact the performance of machine learning models generated to predict WSN data faults. Such as the performance results of one of the classification algorithms used, Support Vector Machine (SVM), the performance of the resulting model on the Accuracy measurement parameter has a value range between 0.29 to 0.83, depending on the model parameters used. In comparison, the Extra- Tree algorithm generates the best model performance on the Accuracy measurement parameter of 96% on all models with the model parameters used.