Using Markov Chain Monte Carlo Algorithm for Sampling Imbalance Binary IDS Datasets
Najmeh Abedzadeh, Matthew Jacobs · 2022
Malicious cyber-attacks can hide themselves in large imbalanced datasets which makes it difficult for Intrusion Detection Systems (IDS) to detect them. Much research has been done in using machine learning and deep learning algorithms for Intrusion Detection Systems, but only a small number of them address imbalanced datasets. In this paper, we apply Markov Chain Monte Carlo (MCMC) Algorithm, Generative Adversarial Networks (GANs) algorithm, and oversampling to the CSE-CIC-IDS2018 dataset to balance the dataset. Then, we compare different Machine Learning algorithms with original dataset, balanced dataset with oversampling, balanced dataset with GAN, and balanced dataset with MCMC. The results show that neither MCMC Algorithm nor GANs is a good algorithm for balancing imbalance binary IDS datasets. The results show that Logistic Regression with original imbalanced dataset provided the best performance in predicting the attacks in imbalance binary IDS datasets with accuracy of 0.88, precision of 0.88, recall of 0.99, Mean Error of 0.11, Root Mean Square Error of 0.33, and Mean Absolute Error of 0.11. Neural Network and GANs performance were close to Logistic Regression specifically regarding accuracy, but Logistic Regression was faster.