EFFECTS OF FEATURE SELECTION TECHNIQUES IN DETECTING GRAY HOLE ATTACKS IN AD-HOC WIRELESS SENSOR NETWORKS USING SUPERVISED MACHINE LEARNING ALGORITHMS

Wireless Sensor networks (WSNs) have become immensely popular due to their simplicity, low cost, ease of deployment, and wide application area.WSNs are a group of tiny autonomous sensor-equipped devices that are deployed in physical or environmental conditions for information gathering. Some of the applications of WSN are forest fire detection, the Establishment of smart roads, tracking parking zones, etc. WSN introduces numerous security threads as a result of its widespread use. The most frequent attack that can harm WSNs is a DOS attack. Grayhole attack is one of the popular attacks against WSNs. The Gray Hole attack is extremely harmful to sensor node networks and causes widespread network malfunctions as well as communication issues across all sensor networks. In this paper, a security mechanism is been proposed to detect grayhole attacks in WSN using the Machine Learning model. Moreover, multi-class classification has been performed on the WSN-DS dataset aiming for gray hole attack detection, 3 different feature selection techniques have been performed and 7 different supervised machine learning classifiers have been implemented on the 3 different feature sets. Parameters such as Precision, Accuracy, Recall, F1-support, and validation are used for evaluation and comparison purposes.Out of all the feature selection techniques, the Univariate Statistical method performed the best with the highest accuracy of 99.8% in the RFC and DTC model.