Amalgamation of K-means Clustering Algorithm with Standard MLP and SVM Based Neural Networks to Implement Network Intrusion Detection System

Intrusion Detection Systems (IDS) are becoming an essential component usually in network and data security weapon store. Since huge amount of existing off-line data and newly appearing network records that needs analysis, data mining techniques play a vital role in development of IDS. The key idea of using data mining techniques for IDS is to aim at taking benefit of classification capability of supervised learning based neural networks and clustering abilities of unsupervised learning based neural networks. In this paper, we propose an efficient intrusion detection model by amalgamating competent data mining techniques such as K-means clustering, Multilayer layer perception (MLP) neural network and support vector machine (SVM), which significantly improve the prediction of network intrusions. Since the number of clusters desired for intrusion detection problem is defined by user a priori and does not change, we employed K-means clustering technique. In the final stage, SVM classifier is used as it produces superior results for binary classification while compared to the other classifiers. We have received the best results and these are compared with results of other existing methods to prove the effectiveness of our model.