Applying Machine Learning and Deep Learning Techniques for Intrusion Detection Systems.

Abstract

Network security risks have significantly increased in recent years due to the rapid growth of digital technology, leading to the emergence of new and advanced threats. To keep pace with these developments, experts are compelled to adopt artificial intelligence (AI)-based solutions to ensure robust defense systems, thereby enhancing their ability to protect networks from escalating cyber threats.This project leverages Machine Learning (ML) and Deep Learning (DL) techniques to enhance Intrusion Detection Systems (IDS), which are considered the frontline defense against cyber threats. Our primary objective was to develop and evaluate the performance of the trained ML/DL models using well-known metrics. As a foundation, we began with a descriptive and comparative analysis of various IDS-related datasets. Based on this analysis, we selected the CIC-IDS2017 dataset for our study. We then implemented a robust data preprocessing methodology, inspired by best practices established in related works.The CIC-IDS2017 dataset was used to train our ML/DL models, achieving robust performance across most evaluation metrics on the test set. However, a critical challenge in IDS design is ensuring model resilience against unseen traffic. To address this, we extended our evaluation by validating our trained models on the CIC-IDS2018 dataset, which contains different attack vectors and network configurations. The models demonstrated strong generalization capabilities, maintaining high performance on this unseen data.The results of our study demonstrated the significant potential of ML/DL techniques in enhancing network security.