Fault Detection using hybrid of KF-ELM for Wireless Sensor Networks

The adoption of data aggregation depending on data fusion and data acquisition for wireless sensor networks (WSN) is increasing these days. While in WSN, the sensor node senses data and send them to the end node. The application of WSN gets limited due to its features such as low-cost sensor nodes, limited battery backups. The usage of sensor nodes in WSN becomes prone to faulty behavior due to its resource constraint and easily gets defected. Predictive detection using data fusion can be a better choice in order to detect the fault with low transmission energy and low power usage. Considering the conditions of the sensor node with its limited capacity of storing and processing of data, a hybrid predictive classification technique is proposed by using the Kalman filter with Extreme learning machine. Here for data fusion Kalman filter is used to train the sink node with the faulty pattern of data in place of training it with the larger amount. In addition, Extreme learning machine (ELM) is used as a predictive classifier, which can provide a high prediction with low communication overhead. The proposed work is evaluated using standard WSN data by inserting random anomalies to it. The performance is measured in terms of detection accuracy and computational time.