Abstract
Nowadays, wireless sensor network (WSNs) becomes an essential device to collect data from the large target phenomenon, especially for continuous environmental monitoring. As the sensor nodes are deployed at harsh and difficult domain area, sensor nodes are prone to malicious and unintended
attack, unexpected device failure or unusual phenomenon event. Therefore, raw data sensed by sensor nodes are inaccurate and not reliable for decision-making and need further processing. As sensor nodes possess to resources constraint in term of energy, processing, and storage anomaly detection
technique must design in a lightweight manner. In our proposed anomaly detection technique, we have incorporated one-class support vector machine (SVM) formulation called Centered Ellipsoid SVM (CESVM) with Candid Covariance-Free Incremental Principle Component Analysis (CCIPCA) for lightweight
anomaly detection. Real sensor data from Sensorscope system project have been tested in term of effectiveness and efficiency of the proposed model. The experiment shows CESVM-DR techniques result in better compared with CESVM.