Multi-class fault type classification in wireless sensor networks using supervised learning on simulated data

Wireless Sensor Networks (WSNs) are used in many vital domains such as environmental monitoring, industrial automation and smart cities. In these areas, dependable data transfer and constant operation are required. However, the existing fault detection techniques in WSNs offer mainly binary classifications; i.e., fault or fault-free without identification of the type of the fault. This is adding to inefficient and delayed troubleshooting and recovery. This work proposes a multi-class supervised machine learning-based fault classification system that can be used to identify five different network conditions: normal, no signal, high packet loss, poor Signal-to-Noise Ratio, SNR, and congestion-induced delay. To this end, the proposed system simulates different failure and recovery situations in the form of Cisco Packet Tracer by measuring important parameters such as RSSI, Packet Loss, SNR and end-to-end delay. Then, the dataset labelled is used to make a Decision Tree Classifier with accuracy above 90%. This work proposes for interpretable, light weight multi-class fault diagnoses for educational and operational improvements in WSNs.