Performance-Based Feature Selection Using Decision Tree

Good, accurate and efficient classification results are expected from a good intrusion system. Intrusion Detection System (IDS) performance depends on multiple factors like accuracy, false positive rate, precision, recall, and f-score, also capability to detect new and stealthy attacks. An IDS must have this capability. In this paper, to achieve highly accurate results, the proposed approach focuses on feature selection using training functions. The data set used for experimentation is CICIDS-2017(Friday). However, the time required to train and test the model is very high and also accuracy is comparably less. To address this, this paper proposed a machine learning classifier based feature selection approach, in which features are selected by considering training time, testing time and accuracy of the model built. In this proposed approach original 82 feature size is reduced to 15 features with providing good accuracy than earlier in less time.