A novel ensemble learning system for cyberattack classification [Póster]

Abstract

This article introduces a novel approach to en hancing cybersecurity through AI by analyzing network traffic, proposing a two-stage cyberattack classification model to handle class imbalance with a one-vs-rest strategy. It aims to differentiate between legitimate and illegitimate network traffic, employing binary models in the first phase to separate traffic types, and an ensemble model in the second phase for detailed classification. Utilizing the UNSW-NB15 dataset for performance evaluation, the proposed system demonstrates superior results, achieving an F1 score of 0.912 in binary classification and 0.7754 in multiclass classification, outperforming other contemporary methods by 0.75% and 3.54% respectively in F1 score.