An Efficient Data Processing Scheme for Wireless Sensor Network Monitoring Using a Machine Learning Model

Wireless sensor networks (WSNs) are playing an increasingly important role in monitoring massive sensors to precisely detect anomalous phenomena, including anomalous events and sensor data faults. Prior studies preferred to dig the event anomaly (e.g., hotspots in a room), while sensor data faults were simply regarded as noise. Considering that different anomalies arise for different reasons, some substantial hidden problems such as internal sensor failures may be ignored. In this study, we propose an efficient data processing scheme using machine learning model with the objective of achieving satisfactory anomaly detection performance during WSN monitoring. Our proposal analyzes the difficulty of detecting different types of fault data and the influence of each type on event detection results. The machine learning model is adopted to analyze the sensor data correlation, to achieve satisfactory performance for both event detection and fault detection by analyzing the correlated sensor data. At each monitoring time during the data monitoring process, the trivial sensor data faults that might affect the event detection results are filtered out before executing event detection. Meanwhile, at much longer monitoring time intervals, random fault detection is performed to find potentially hidden failures of sensors. Numerical experiments conducted in a real WSN environment show that neural network model outperforms other machine learning models in anomaly detection, and the results by adopting neural network model verify the feasibility of our proposed scheme which attains acceptable performance in detecting both types of anomalies.