Intrusion Detection System using Ensemble based Feature Selection Technique

In recent years, the activity of internet devices and software has drastically increased hence, the security of networks has become a serious concern. To tackle this concern, an Intrusion detection system (IDS) plays a pivotal role. Enhancing the performance of intrusion detection systems involves employing feature selection techniques to identify a relevant subset of features within the dataset. This approach not only diminishes computational time but also enhances the accuracy of the IDS. It's crucial to check that the selected features are both important and able to be predicted well. This ensures the intrusion detection system (IDS) works effectively. In this research paper, we introduce an ensemble feature selection method for constructing an IDS. Gain ratio and Spearman's correlation coefficient are used in the ensembled method for feature selection. To implement this method, the CICIDS-2017 dataset is used. The top-ranked features are selected using the proposed method, and then an arithmetic operation is performed on two sets of generated features. This specific set of selected attributes is subsequently given to the machine learning (ML) classifier for training and testing. Among the evaluated classifiers, random forest has shown the best result. It has shown the accuracy of 99.78%.