A Study of Imbalanced Dataset Classification on KDD99 Datasets with Reinforcement Learning Mechanism
Chih-Chen Pan, Yungho Leu · 2023
In recent years, the application of machine learning to Network Intrusion Detection Systems (NIDS) has emerged as a crucial area of research. While NIDS is critical in safeguarding our networks against cyber-attacks, it still faces limitations in detecting unknown attacks, such as 0-day attacks[1]. Furthermore, current NIDS technologies need high prediction accuracy, high recall rates, and low false alarm rates. However, imbalanced datasets for NIDS have resulted in low recall rates. To address this issue, we propose a reinforcement learning algorithm-based mechanism (RLAM) to improve the performance of network-based intrusion detection on imbalanced datasets. We verify the effectiveness of RLAM by applying it to improving the performance of the Neural Network and Decision Tree. Experimental results demonstrated that RLAM significantly improved the recall rate of a critical-and-important (CAI) minor class (U2R) in KDD-CUP99 datasets without sacrificing the performance measures of the other classes. Notably, the enhanced decision tree algorithm outperformed all the existing intrusion detection algorithms in terms of overall accuracy.