Advancing Intrusion Detection with Machine Learning: Insights from the UNSW-NB15 Dataset
Atul Kumar, Kalpna Guleria, Rahul Chauhan, Deepak Upadhyay
Abstract
This research undertakes a comparative examination of the recall performance of Logistic Regression (LR), Random Forest (RF), and Decision Tree (DT) models for intrusion detection systems (IDS). The analysis is conducted using the UNSW-NB15 dataset, which is a comprehensive compilation of contemporary cyber-attack scenarios. The objective is to assess the effectiveness of these models in their respective roles. According to our investigation, Random Forest has superior performance compared to LR and DT in the detection of network intrusions. It exhibits the greatest recall values, which are of utmost importance in mitigating false negatives in Intrusion Detection Systems (IDS). The exceptional efficacy of RF highlights its resilience and versatility in addressing the intricate and diverse array of cyber threats. While LR demonstrated impressive outcomes, particularly when dealing with less complex assault methods, and DT provided simplicity in interpretation, both were somewhat less efficient when compared to RF. The results indicate that Random Forest has great potential for use in modern Intrusion Detection Systems (IDS), prompting more investigation into ensemble methods, deep learning, and unsupervised learning approaches to improve threat detection. This work contributes to the progress of machine learning-based security measures and guides the creation of more efficient and adaptable Intrusion Detection Systems (IDS) that can effectively address the changing nature of cyber threats.