CYBER SECURITY MACHINE LEARNING MODEL ASSESSMENT: PERSPECTIVES FROM THE UNSW-NB15 DATASET

Objective: Detection of cyberattacks still remains as one if the challenges. Method: The paper compares the performance of five machine learning classifiers namely Decision Tree (DT), XGBoost (XGB), Gaussian Naive Bayes (GNB), Random Forest (RF) and Logistic Regression (LR)) with respect to classifying network traffic as normal or malicious. It leverages the UNSW-NB15 dataset, which incorporates a wide variety of modern attack types and employs extensive data preprocessing, feature evaluation, and data discussion stratified splitting to ensure robust results. Results: The results show that the top performing model, XGBoost, has an accuracy of 93.62 % and AUC at 0.99 which means A great performance. Random forest and decision tree follow with accuracies of 93.60% and 92.93%, although less cumbersome fashions such as logistic regression and Gaussian Nav Bayes appear less accurate due to their limitations in dealing with complex communities in non-subnetworks traffic structure for real-time records tested. Novelty: This underlines the promise of advanced machine expertise, especially ensemble and tree-based perfection strategies, to improve automated cyber risk detection. Future efforts will focus on feature optimization and exploring streaming data.