Improving Intrusion Detection Accuracy using Convolution Neural Network

Network Intrusion Detection has been an active area of research with the growing incidences of cybercrimes. This has led to continuous monitoring of network traffic, analysis, and raise alarm if any abnormality is noticed so as to trigger appropriate response in order to curb the possibility of an attack. One of the approaches to deal with the network intrusion problem is to classify the network user behavior as normal or suspicious. Soft computing based techniques are being tried out to classify network users with higher degree of accuracy and low false alarm rate. In this paper, we propose a classification model for the detection of known as well as unknown network attacks based on artificial neural network based techniques namely, RBFN, SOM, LVQ3, SMO, and CNN. Further, in order to improve the performance of the classifier, Z-Score normalization has been applied for preprocessing of data. The performance of the model has been evaluated on the NSL-KDD dataset in terms of Precision, Accuracy, Detection rate, F-Value, and False Alarm rate. Keywords: Convolution Neural Network, Learning Vector Quantization, Normalization, Self-Organizing Map, Sequential Minimal Optimization, and Radial Basis Function Network.