Entropy Feature Based Outlier Detection using ANN for WSN Application
Tanya Singh · International Journal of Advanced Trends in Computer Science and Engineering · 2020
Wireless sensor networks (WSNs) are composed of a large number of tiny sensor nodes deployed in an environment for monitoring and tracking purposes.Sensor nodes use ad-hoc communications and collaborate with each other to sense different phenomena that may vary in time and space, and send the sensed data to a central node for further processing and analysis.An anomaly is an observation in a data set, which appears to be inconsistent with the remainder of that data set.The dynamic environment of network and roughness of the working condition are also responsible to generate inaccuracy in measurements.In this paper, an approach for outliers detection based entropy value of received sensor voltages is applied using ANN prediction model .The algorithm development and analysis involves a real time database generated on 14 sets of MICA2 wireless sensor kit with anomaly inserted by real time motion based intrusion in the lab by volunteers from Intel Berkeley lab.On each sensor data pair segmentation is applied by fixed window size in order get large outliers' measurements training dataset.The analysis demonstrates the measurement accuracy in detection of number of outliers that its 86%.Moreover, the algorithm also provides an analysis in terms of impact of variation in learning types and number of nodes in the ANN prediction model.This work is helpful in the application in the situations where high amount of noise or distortions are present.The outlier part from distorted data can be figured out and recollected to enhance application accuracy.