ABSTRACT

With the development of web the world has changed into a worldwide advertisement platform with all financial and business practices being conveyed on the web. Being the most basic asset of the creating scene, it is the helpless protest and thus should be secured from the clients with perilous identity set. Since the Internet does not have central observation segment, attackers once in a while, using different progressive hacking topologies find a way to sidestep frameworks security and one such gathering of attacks is Intrusion. An intrusion is a development of breaking into the structure by trading off the security game plans of the structural set up. The strategy of taking a gander at the framework data for the possible intrusion is known to be intrusion detection. Throughout the previous two decades, programmed intrusion discovery framework has been an essential point of thorough investigation. Till now scientists have created Intrusion Detection Systems (IDS) with the ability of recognizing assaults in a few accessible situations; most recent on the scene are Machine Learning approaches. In this paper, the preprocessing unsupervised discretization and feature selection method has been applied to enhance the classification accuracy. Unsupervised discretization method is extremely important to make NSL-KDD data set as appropriate input for experiment. The discretization method needs to map these non numeric values into numeric values of features for helping classifier. After discretized the data, Principal Component Analysis (PCA) method is applied to generate subset features from whole data set. PCA method has been used to reduce dimensionality from dataset. The Naïve Bayes classifier has been employed for classification data as synonymous attacks or normal. For experimental analysis, the NSLKDD standard data has been used to evaluate the proposed model. The empirical analysis results of proposed model demonstrate that it is better in terms of all performance measures. A comparative analysis of the results obtained for the proposed model using preprocessing methods and existing NaiveBayes algorithm with original data is presented. The empirical results prove that the performance of the proposed model is more robust and better than the performance of existing method. Keywords; IDS, NSL-KDD, proposed model, R2L and U2R attacks

https://ijbpas.com/pdf/2018/April/MS_IJBPAS_2018_44073.pdf

Download PDF