An Effective Hybrid Feature Selection and Classifier Model for Intrusion Detection Systems (IDS)

Abstract

Network intrusion threats can have an impact on business losses. One mechanism that can be applied is Intrusion Detection Systems (IDS). IDS model development is carried out in various categories, starting from experimenting with classifiers, combining classifier models, and carrying out optimization, including implementing various feature selections. This experiment cannot be separated from the current need for an accurate IDS model with a minimum response time. This study was conducted to find the most efficient and proper combination of classifiers and features from three feature options, namely Grey Wolf Optimizer (GWO), Gain Ratio, and Chi-Square. At the same time, the classification algorithms used are Logistic Regression, Support Vector Machine, Random Forest and Decision Tree. The combination of these models will be tested on the NSL-KDD and UKM-IDS20 datasets. The tests showed that the Random Forest classifier can be used hybrid with the GWO feature selection algorithm and produces high accuracy with low computation time. In detail, for the NSL KDD dataset, the combined GWO-RF model has the highest accuracy, with 99.99% for training and 99.89% for testing. The GWO-RF model outperformed all other feature selection and classifier alternatives on the UKM-IDS20 dataset in terms of accuracy, where the resulting accuracy value reaches 99.98% for training and 99.97% accuracy for the testing process.