Al-Farg'oniy avlodlari, 2026, № 3, с. 310-317
Cyberattack Detection in Network Traffic: A Hybrid Ensemble of Gradient Boosting and LSTM on the CSE-CIC-IDS2018 Dataset
Аннотация · Abstract
Signature-based intrusion detection systems fail to recognize new attacks, and rare attack classes remain difficult for machine learning models. This study proposes a hybrid ensemble that combines XGBoost, LightGBM, CatBoost, a deep neural network and an LSTM network through soft voting with Bayesian-optimized weights. On the CSE-CIC-IDS2018 dataset (15 classes), the ensemble achieved 98.84% accuracy and a macro-F1 of 0.7924, compared with 0.7019 for the best individual model (Random Forest), while the difference in accuracy was negligible. Removing the LSTM reduced macro-F1 to 0.6781. The infiltration class remained the main source of errors.
Сведения · Details
- Журнал · Journal Al-Farg'oniy avlodlari
- Язык · Language английский
- Цитирования (Crossref) · Citations 0