Network intrusion detection leveraging machine learning and feature selection

Abstract

Handling superfluous and insignificant features in high-dimension data sets incidents led to a long-term demand for system anomaly detection. Ignoring such elements with spectral instruction not speeds up the analysis process but again facilitates classifiers to make accurate selections during attack perception stage, when wrestling with huge-scale and heterogeneous data. In this paper, for dimensionality reduction of data, we use Correlation-based Feature Selection (CFS) and Naïve Bayes (NB) classifier techniques. The proposed Intrusion Detection System (IDS) classifies attacks using a Multilayer Perceptron (MLP) and Instance-Based Learning algorithm (IBK). The accuracy of the introduced IDS is 99.87% and 99.82% with only 5 and 3 features out of 78 features for IBK. Other metrics such as precision, Recall, F-measure, and Receiver Operating Curve (ROC) also confirm the principal performance of IBK compared to MLP.