ROC Based Performance Evaluation of Machine Learning Classifiers for Multiclass Imbalanced Intrusion Detection Dataset
Md. Salman Mohosheu, Fatin Abrar Shams, Md. Abdullah Al Noman, Samiur Rashid Abir, Al-Amin · 2023
Distinguishing between hazardous and benign traffic is a crucial issue in network security. Strategies like firewalls that restrict unauthorized network access using established rules and Artificial Intelligence are already in use. This study employs various supervised machine learning methods to classify malicious network traffic. Furthermore, this study has determined that accuracy and False Acceptance Rate are not particularly informative metrics in this context because of the highly imbalanced nature of the dataset utilized in this research. More appropriate metrics, such as the F1 score and ROC score, are recommended. ROC analysis revealed that for complete coverage, accurate hostile traffic classification necessitates allowing a 30% False Positive Rate. Among all the algorithms tested in this work, the Random Forest algorithm exhibited the highest performance with an F1 score of 0.81, while XGBoost achieved the highest ROC score of 0.9695. With thorough research, Machine Learning techniques can be used to mitigate cyberattacks successfully.