Anomaly Detection in Wireless Sensor Networks Using Self-Organizing Map and Wavelets
Source: NORTH ATLANTIC UNIVERSITY UNION
This paper proposes an anomaly detection scheme which is able to detect anomalies accurately by employing only important features of data signals, instead of using all the sensor data traces. The contribution of this paper centers on anomaly detection by using Discrete Wavelet Transform (DWT) combined with a competitive learning neural network called Self-Organizing Map (SOM) in order to accurately detect abnormal data readings while using just half of the data size. Experiment results from synthetic and real data injected with synthetic faults collected from a WSN show that the proposed algorithm outperforms the SOM algorithm by up to 18% and DWT algorithm by up to 35% in presence of bursty faults with marginal increase of false alarm rate.