Machine Learning Techniques Applied to Intrusion Detection Systems

As cyber threats continue to evolve, the necessity of Intrusion Detection Systems (IDS) to protect sensitive information has become increasingly apparent. To address the limitations of conventional signature-based IDS, machine learning-based approaches have been proposed. However, the dearth of up-to-date network datasets impedes the advancement of these innovative methods. Furthermore, robust and reproducible results are required to ensure that the performance of ML models is not merely a product of randomness but a reflection of genuine and reliable capabilities. In this context, this paper presents a comprehensive evaluation of standalone and hybrid machine learning algorithms aimed at improving the differentiation between normal and malicious traffic. The hybrid model combines rule-based filtering with machine learning-based evaluation. A set of predefined rules is first applied to filter the dataset, and then the ML algorithms focus on the most relevant data, thus improving the efficiency in detecting previously unseen threats and reducing computational complexity. This approach is designed to improve detection accuracy while maintaining a low false positive rate, addressing both precision and operational efficiency in real-world scenarios.