Applying Gini Importance and RFE Methods for Feature Selection in Shallow Learning Models for Implementing Effective Intrusion Detection System
Nilesh G. Pardeshi, Dipak V. Patil · Advances in computer science research · 2023
Cyber security is becoming important concern in the recent world.Number of internet users are increasing day by day and they are accessing huge amount of data on their device from different websites.Attackers are trying to get access to normal user's systems by introducing different types of attacks.Number of Intrusion Detection Systems are being developed to protect the normal users from the attackers.Most of these systems are developed using outdated datasets, and they are having scope to improve their accuracy, detection rate and to reduce false alarm rate.In the proposed system we are going to train the different machine learning models for binary and multiclass classification.We are using shallow learning algorithms like Decision Tree, Random Forest, Naïve Bayes, K-Nearest Neighbor, Support Vector Machine, XGBoost and Ensemble Technique on benchmark NSL-KDD and recent CICIDS-2017 IDS dataset.We have used Recursive Feature Elimination and Feature Importance based features selection on NSL-KDD and Feature Importance based feature selection on CICIDS-2017 dataset.All machine learning models are trained using all features and selected features datasets.Testing is performed on separate test dataset like KDDTest+ as well as test sets obtained by applying train test split on original datasets.It is observed that the performance on feature importance-based feature selection models and 10-fold cross validation models is improved in terms of accuracy, precision, recall and f-measure.