Intrusion Detection Using PCA Based Modular Neural Network

Abstract

Most of current intrusion detection systems are based on machine learning methods but very few till now use clustering algorithms as a preprocessing layer to reduce the high dimensionality of data, which is difficult to analyze. In this paper we introduce Modular Neural Network for intrusion detection, which apply Principal Component Analysis (PCA) as preprocessing layer for reducing huge information quantity presented in knowledge discovery and data mining (KDD99) data set. PCA significantly reduce the high dimensionality of data set without loss of information. Then this preprocess data in the form of principal component is presented to Batch Backpropagation Neural Network for efficient intrusion detection. We rely on some experiments to calculate Root Mean Square Error (RMSE) using Modular Neural Network on KDD 99 data set. Our experimental results show improvement in the learning time due to the reduction of high dimensions of data. Also we have obtained low RMSE during training, which is below the acceptance range of 0.1. Proposed Modular Neural Network has capability to efficiently and accurately classify data into attack and normal.