Advancing Cybersecurity: Leveraging UNSW_NB15 Dataset for Enhanced Detection and Prediction of Diverse Cyber Threats

Abstract

A significant challenge in this domain is the lack of a complete network-based dataset that accurately represents contemporary network traffic patterns, encompasses a wide range of subtle intrusions, and provides detailed, structured information on network activity. Current Intrusion Detection Systems (IDS) struggle with high rates of both false positives and false negatives, leading to reduced accuracy in detecting a wide range of cyber threats. This inconsistency affects the overall effectiveness of these systems in identifying diverse types of attacks. In this paper we discuss the UNSW-NB15 dataset serves as a prominent benchmark for research in network intrusion detection. It was designed to support the advancement and assessment of intrusion detection systems (IDS) and machine learning techniques aimed at identifying and categorizing different forms of network attacks.This study focuses on developing machine learning model that can identify cyber-attacks and and enhance IDS system performance. The UNSW_NB15 dataset contains 9 different types of cyber-attacks namely Fuzzers, analysis, DoS, Backdoor, Exploits, Generic, Reconnaissance, Shellcode, worms. We obtained the dataset from link - http://www.cybersecurity.unsw.adfa.edu.au/ADFA%20NB15%20Datasets/.

The proposed model was executed with different algorithms such as logistics regression Nearest Neighbour, Decision Tree and after analyzing the results, we observed that the Random Forest algorithm achieved remarkable accuracy, precision, recall, and F1 score of 98%, closely followed by the Decision Tree algorithm with an impressive 97% accuracy. However, the Support Vector Machine (SVM) algorithm demonstrated relatively lower accuracy at 54%.