Anamoly Detection in Wireless Sensor Networks

In a sensor node outliers tend to occur by chance or by attacks. These occurrence of outliers can effect the reliability of the process that the sensor node handles. Outliers are defined as those data which differ from the normal behaviour of data. Over the years, the demand for outlier detection has increased significantly. Any detection scheme used should consider many constraints such as communication overhead, energy and computational complexity. The paper proposes to compare methods based on correlation such as one class support vector machine, a mixed algorithm of K-means clustering along with compression techniques and when high dimensional dataset is used, a local outlier mining approach namely LOMA for mining outliers in an efficient way. Spaciotemporal correlation drives concept of SVM and attribute relevance helps to mine in LOMA. Experimental results compares the effectiveness of the methods in detection of outliers in respective platforms.