Improved Detection of Probe Request Attacks - Using Neural Networks and Genetic Algorithm

Abstract

The Media Access Control (MAC) layer of the wireless protocol, Institute of Electrical and Electronics Engineers (IEEE) 802.11, is based on the exchange of request and response messages. Probe Request Flooding Attacks (PRFA) are devised based on this design flaw to reduce network performance or prevent legitimate users from accessing network resources. The vulnerability is amplified due to clear beacon, probe request and probe response frames. The research is to detect PRFA of Wireless Local Area Networks (WLAN) using a Supervised Feedforward Neural Network (NN). The NN converged outstandingly with train, valid, test sample percentages 70, 15, 15 and hidden neurons 20. The effectiveness of an Intruder Detection System depends on its prediction accuracy. This paper presents optimisation of the NN using Genetic Algorithms (GA). GAs sought to maximise the performance of the model based on Linear Regression (R) and generated R > 0.95. Novelty of this research lies in the fact that the NN accepts user and attacker training data captured separately. Hence, security administrators do not have to perform the painstaking task of manually identifying individual frames for labelling prior training. The GA provides a reliable NN model and recognises the behaviour of the NN for diverse configurations.