Comparative Study of Anomaly Detection in Wireless Sensor Networks Using Different Kernel Functions

Wireless sensor network (WSN) is defined as an autonomous network composed of low power sensor nodes having limited computational, communication, and energy resources. Being short at resources they require efficient use of each resource to prolong network lifetime. Sensor networks are exposed to noise, compromised nodes, low battery levels, and damaged sensors, all these results in anomalous readings or anomaly. Presence of anomaly in system deteriorates the performance of WSN in terms of efficiency, accuracy, and reliability. Hence anomaly detection becomes a major challenge to decide the performance of network. Support vector machine (SVM) is a light weight, learning-based binary classifier that can classify the raw data into normal and anomalous. SVM suffers from computational complexity while handling large datasets, so sequential minimal optimization SVM (SMO-SVM) is used to reduce the complexity. In this paper, a comparative study is made on anomaly detection using SMO-SVM classifier utilizing different kernel functions.