Anomaly Detection in Wireless Sensor Networks Using S-Transform in Combination with SVM

In this paper, we propose a novel method of anomaly detection in wireless sensor networks (WSN) based on S Transform. It makes use of S transform for feature extraction. We extract only the significant components of the time-series data. Earlier wavelets based approach that extracts features from the time-series data has been applied for detecting anomalies in combination with various classifiers like Self Organizing Maps and SVM (Support Vector Machine). The wavelet based approach considerably reduces the processing time of Anomaly Detection System (ADS) as the data size is considerably reduced. However these methods are characterized by high computational complexity. It has been observed that the S-transform could give better frequency resolution than wavelet based approaches. The application of S-Transform to the time series data considerably reduces the data size. Therefore, the proposed method does not makes use of huge amount of data in processing the information sought, and hence can efficiently detect and classify different types of fault with little processing time. It aims at detecting and classifying anomalies at node level according to the characteristics of data collected by each individual sensor. These features obtained from S-transform when integrated with SVM can classify data into class 1(Original Signal) and class 2(Anomalous Signal). It has been observed that the method give higher accuracy with lesser computational complexity.