Survei Teknik Pemilihan Fitur Untuk Sistem Deteksi Intrusi Berbasis Machine Learning

Abstract

With the increasing threat to cybersecurity, Machine Learning (ML)-based Intrusion Detection Systems (IDS) are becoming increasingly important for detecting and preventing network attacks. The selection of appropriate features is a key factor in improving the performance of IDS, as it can enhance detection accuracy, reduce model complexity, and save computation time. This article examines various feature selection techniques used in ML-based IDS, including filter, wrapper, embedded, and hybrid techniques. Each technique has its advantages and disadvantages, depending on the characteristics of the dataset and the type of attack encountered. This research also evaluates the effectiveness of these techniques using popular datasets such as KDD Cup 99, NSL-KDD, and CICIDS 2017. The results show that filter techniques are more efficient in terms of time, while wrapper and hybrid techniques offer higher detection accuracy, although they require more resources. The embedded technique combines efficiency and accuracy with time savings in model training. This article also discusses the importance of good feature selection for classification in IDS, as well as the challenges faced by IDS in overcoming its limitations. This research provides a comprehensive overview of feature selection in ML-based IDS and recommendations for further development and implementation to address increasingly complex threats.