Comparative Analysis of Various Machine Learning Techniques Applied Towards Intrusion Detection in Computer Networks

Abstract

The paper discusses the development of intrusion detection systems (IDS) and their limitations in accurately detecting minority attack classes in computer networks. Despite advancements in IDS technologies, attackers can still breach networks. The aim of the work is to compare various machine learning models to find the best performing one for intrusion detection. The methodology involves using the Boruta algorithm for feature selection, under sampling to address class imbalance, and PyCaret for model comparison, training, and testing. The experimental results reveal that the Gradient Boosting classifier achieved the highest accuracy at 99.70%, while Naïve Bayes had the lowest accuracy at 84.77%. These findings underscore the importance of selecting robust machine learning approaches to enhance network security against evolving cyber threats. A stacking classifier was also created and outperformed other algorithms with 99.69% accuracy but slightly below the Gradient Boosting Classifier, which had 99.72% accuracy. The recommended model of choice for network intrusion detection is the Gradient Boosting classifier.