COMPARATIVE ANALYSIS OF CLASSIFICATION OF MACHINE LEARNING MODELS ON KDDCUP99 AND CIC-IDS DATASETS FOR THE INTRUSION DETECTION SYSTEM.

Abstract

Intrusion Detection System (IDS) is the key part of system defense for growing attacks.  The inconsistent and unreliable performance of anomaly-based intrusion detection methods can be attributed to outdated test and validation datasets. This study investigates the performance of various machine learning models on the KDDCup99 and CIC-IDS popular datasets, focusing on their predictive performance accuracy, computational efficiency and generalizability. A suite of models including logistic regression, decision trees, random forest, support vector machines, gradient boosting were evaluated. Results indicate that different ML models and deep learning methods outperform traditional algorithms in terms of accuracy, but they require higher computational resources. Model performance varied significantly between the two datasets, highlighting the impact of data characteristics on model efficiency.