A systematic literature study of machine learning techniques based intrusion detection: datasets, models, challenges, and future directions

The tremendous growth of cyberspace has magnified the risk of advanced cyber-attacks, making the conventional security system deficient. An intrusion detection system (IDS) is a crucial aspect for securing the networks, and machine learning based approaches have attained significance for detecting growing threats, which is extremely essential to address the major concern of network security. This survey paper systematically reviews the machine learning-based IDS, focusing on detection models, the most used datasets, and evaluation metrics. A systematic review methodology, including defined selection criteria and a detailed analysis framework, enables clarity and reproducibility. Furthermore, the paper presents a comparative analysis of popular IDS datasets such as NSL-KDD, CICIS2017, and UNSW-NB15 and points out their limitations, including imbalance and antiquated attack methods. The key contributions are: (a) an organized taxonomy of ML-based IDS techniques, (b) a detailed comparison of datasets and related challenges, and (c) pointing out the research gaps to guide the future directions. This survey paper aims to help scholars in developing vigorous, intelligent, and convertible intrusion detection solutions.