CRAMER'S CORRELATED DEEP CONVOLUTIONAL AND COX REGRESSIVE BOOTSTRAP CLASSIFIER FOR DOS ATTACK DETECTION

Abstract

Wireless sensor network (WSN) occupies a certain role in diverse applications that needs data collection and transmission. With the random node deployment, they are susceptible to attacks. Most frequent attacks on WSNs which goal every layers of protocol stack is DoS attack. This attack directly affects network performance as well as prone network security. In this paper, a novel method called Cramer's correlated deep convolutional and Cox regressive Bootstrap Classifier (CCDC-CRBC) is proposed for DoS attack detection in WSN with enhanced accuracy, minimum time. CCDC-CRBC method comprises three different steps to determine the different types of DoS attacks in the network. First, a Z-score normalization-based preprocessing model is employed to eliminate the noise in the input dataset.  Then the normalized outputs are fed to the process of feature extraction for selecting pertinent features. The process of feature extraction is performed by using Cramer's correlated deep convolutional learning. The deep convolution network comprises several layers to accurately learn the given input and extract more pertinent features. With the extracted features, the classification of different kinds of DoS attacks is made using Czekanowski Cox Regressive Bootstrap Aggregative Classifier. Simulation evaluation of the CCDC-CRBC method is performed by different metrics namely attack detection accuracy, attack detection time, precision, recall, f-measure. The outcomes demonstrate CCDC-CRBC method efficiently   improves performance of DoS attack detection in WSN.