Approach to Improving Machine Learning Models for Intrusion Detection System

Abstract

In today's digital world, there are several security risks that digital assets must contend with. Systems for detecting intrusions (IDS) are essential security tools that protect digital assets. But their usefulness depends on meeting strict accuracy requirements, and their effectiveness depends on timely alarms. This study offers a novel IDS model that combines deep learning and machine learning methods as a solution to these problems. The study applies several classification techniques, such as Gaussian Naive Bayes (GNB), Random Forest (RF), Decision Tree, K-Nearest Neighbors (KNN), Soft Voting, and Hard Voting, using the well known KDD Cup-1999 dataset. After a large-scale dataset was processed, the Decision Tree method performed better than the others, with a 99.9% accuracy rate. This study aims to investigate the effects of soft voting and hard voting, a novel application in IDS. Decision Tree proved to be the better performance in spite of these efforts. By offering information about algorithmic efficacy, the research advances the field of intrusion detection and helps decision-makers in the design and deployment of intrusion detection systems. These findings have implications for improving digital asset protection against changing cyber threats.