One class SVMs outlier detection for wireless sensor networks in harsh environments: Analysis

Abstract

Outlier/Anomaly detection is renewed challenge in data mining, internet of things as well as machine learning communities. In present era Internet of things is emerging with its tremendous applications where wireless sensor nodes are incredibly well structured to accumulate huge amount of raw data from unsystematic sectors and hand over it to authoritative systems such as disaster monitoring, surveillances, green monitoring, and smart city applications etc,. However such authoritative and prediction systems truthfulness subject to reliability of sensor node. Unluckily, sensed data excellence and reliability influenced by circumstances such as sensor faults, intrusion and unusual events within others. As a result it obstructs authoritative decision making as well as prediction, hence there is need of effectual, real time abnormality detection mechanisms for consistent decisions. A key dispute is how to lessen energy consumption and communication overhead in network at the same time identifying anomalies in unsystematic environments. Even though a impressive number of studies, existing anomaly detection algorithms are there still Machine learning numerous appliances has captured massive importance in outlier detection especially for wireless sensor networks (WSNs), notably Support Vector Machine (SVM) based techniques provides effectual outlier detection and classification achievements in harsh environment. This work presents various one class SVM formulations eminently well instructed outlier detection in harsh environments, moreover formulations analyzed in terms of various characteristics include input data, dynamic topology, outlier types, Spatio temporal attribute correlations etc. Brief comparison and characteristics of distinctive one class SVM formulations are described.