Network Intrusion Detection Through Machine Learning With Efficient Feature Selection

Network Intrusion detection is an essential aspect of implementing secure networks. Network Intrusion detection systems help mitigate network attacks preemptively and let the system take preventive measures. Machine Learning techniques enable the system to build attack classification models based on previously available real-world data. Advanced security Network Metrics dataset(ASNM) is one such dataset that contains derived network features that help create a highly accurate network attack classifier. In this paper, we pre-process the ASNM dataset and use feature selection techniques to filter out unwanted features and retain features that are highly correlated to the classification in the ASNM dataset. We apply the Variance threshold and Chi-square Test feature selection techniques on the ASNM dataset. Finally, a neural network model based attack classifier is developed and shown to have a prediction accuracy of 99%.