Security Enhancement in Wireless Sensor Networks using Machine Learning

Ensuring the security of wireless sensor networks (WSNs) is vital for monitoring real-time systems. One of the major security flaws experienced by WSNs is denial of service (DoS) which can even lead to the breakdown of the complete system or to wrong decisions being made by the system that can cause adverse results. This research work focuses on two techniques for detecting a DoS attack at a medium access control (MAC) layer. Our research compares and evaluates the performance of two major machine learning techniques: neural network (NN) and support vector machine (SVM). Vanderbilt Prowler is used for simulating the scenarios. In the simulations, normalized critical parameters and their corresponding probabilities of DoS attack are computed in 50 trial runs. These normalized critical parameters and their corresponding probabilities of DoS attack are used as training inputs in NN and SVM approaches. The simulation results clearly show that SVM provides better accuracy compared to NN, 97% accuracy by SVM and 91% accuracy by NN. The simulation also shows that SVM takes much less time to detect and determine the probability of a DoS attack, 0.25 seconds by SVM and 0.75 seconds by NN. All these results clearly show that SVM performs better than NN when used for detecting the probability of DoS attack in WSNs.