AutoGluon-Based Performance Analysis for Multi-Class Network Attack Detection

Authors : Sinan Kocagöz, Fatih Yücalar, Emin Borandag, Ender Şahinaslan

Pages : 341-350

Doi:10.18586/msufbd.1753107

View : 70 | Download : 125

Publication Date : 2025-12-24

Article Type : Research Paper

Abstract :The increasing complexity and critical nature of cyber threats have heightened the importance of effective attack detection systems. In this study, various machine learning algorithms (SVM, KNN, Logistic Regression, Random Forest, XGBoost), deep learning models (CNN, LSTM, DNN), and the AutoML-based AutoGluon framework are systematically compared for multi-class network attack detection. The experiments utilize the UNSW-NB15 dataset. Due to the imbalanced class distribution in the dataset, class balancing was applied in certain analyses using the SMOTE technique. All models were evaluated using commonly adopted classification metrics, including Accuracy, Precision, Recall, F1-score, and ROC-AUC. The findings indicate that AutoGluon achieved the highest performance, owing to its automated modeling and ensemble-based approach. These results suggest that automated modeling techniques may offer greater competitiveness and effectiveness compared to traditional methods. By systematically analyzing the performance of different modeling strategies in intrusion detection systems, this study aims to provide guidance for the development of future security solutions.
Keywords : AutoGluon, AutoML, Derin Öğrenme, Makine Öğrenmesi, Ağ Saldırısı Tespiti

ORIGINAL ARTICLE URL