Impact of Feature Selection for Intrusion Detection
Abstract
The online platforms and business are generating a lot of data and thus the security of data from attacks is the need of an hour. Researchers are constantly working to develop an intrusion detection method that can improve detection accuracy quickly.However, the presence of large amounts of network data increases the computation time due to presence of redundant and irrelevant features and has a negative impact on the performance of classifiers used in intrusion detection tasks.Feature selection methods plays a vital role in this regard. In this paper, we measure the impact of feature selection on detection performance of machine learning classifiers for intrusion detection using CICIDS2017 dataset.
For this purpose, we used four classifiers namely Logistic Regression, Decision Tree, Naïve Bayes and k-nearest neighbors with all dataset features and feature set obtained by applying Recursive Feature Elimination on CICIDS2017 dataset. The detection performance of classifiers is measured using various metrics like accuracy, recall, precision and f1-score. The obtained results demonstrate that detection performance of machine learning classifiers can be improved significantly using recursive feature elimination for intrusion detection upto 99.98% accuracy.