Outlier detection in WSN by entropy based machine learning approach

Abstract

Environmental disasters like flooding, earthquake, epidemics etc. cause’s significant catastrophic effects on population of all over the world. Wireless sensor network (WSN) based techniques have become significantly popular in susceptibility modelling of such challenging disaster due to their greater strength and efficiency in the prediction of such threats occurring enormously day by day. This paper demonstrates the multiple machine learning-based approach to predict outlier in sensor data records with the use of bagging, boosting, random subspace, SVM and KNN based frameworks for outlier prediction using a Wireless sensor network data records. First of all the algorithm follows the pre processing of the database taken from records of 14 sensor motes with presence of outlier due to intrusion. Subsequently the segmented database is created from sensor pairs. Finally, the data entropy is calculated and used as a feature to determine the presence of outlier used different approach. Results show that the KNN model has the highest prediction capability for outlier assessment.