Outlier Detection and Decision Tree for Wireless Sensor Network Fault Diagnosis

Abstract

Wireless Sensor Network (WSN) has been used in the industrial world and the household. The increasing number of WSN-based smart home devices requires intensive monitoring and automation. Problems may arise when a fault occurs on these devices that result in misinterpretation of the data received. Existing approaches to fault detection and diagnosis have led to the development of fault diagnosis methods for large-scale data. One of the effective methods for fault diagnosis is the Multi-Scale Principal Component Analysis (MSPCA). This research implements a combination of MSPCA and Decision Tree to detect fault data and diagnose the type of fault cause. The classification of faults is based on significant changes in temperature, humidity, light, voltage, as measured from the Normal Profile extracted by the MSPCA. Experiment results showed that our method was able to determine faults with an accuracy score of 0.913.