VAE-GAN-Guided Cross-Class Generation: A Class Imbalance Data Augmentation Method for Network Intrusion Detection

Electronics, Год журнала: 2025, Номер 14(11), С. 2103 - 2103

Опубликована: Май 22, 2025

Network intrusion datasets often face class imbalance issues in detection tasks, where the number of majority samples is much higher than minority samples. Current solutions notable limitations: traditional normalization weakens multimodal distribution continuous features, while mainstream generative models focus excessively on mining neglecting information. To address these issues, we propose M2M-VAEGAN, which innovatively incorporates a Variational Gaussian Mixture (VGM) model to preserve characteristics features. We design transfer learning framework, pre-training classes capture general attack patterns, followed by fine-tuning with balanced batches and prevent catastrophic forgetting. Additionally, enhance VAEGAN architecture an auxiliary classifier strengthen conditional information learning. On NSL-KDD CIC-IDS2017 datasets, M2M-VAEGAN outperforms methods such as SMOTE, CTGAN, CTABGAN, achieving 1.25% 6.42% improvement recall. These results demonstrate effectiveness proposed approach.

Язык: Английский

Securing the Edge: CatBoost Classifier Optimized by the Lyrebird Algorithm to Detect Denial of Service Attacks in Internet of Things-Based Wireless Sensor Networks

Future Internet, Год журнала: 2024, Номер 16(10), С. 381 - 381

Опубликована: Окт. 19, 2024

The security of Wireless Sensor Networks (WSNs) is the utmost importance because their widespread use in various applications. Protecting WSNs from harmful activity a vital function intrusion detection systems (IDSs). An innovative approach to WSN (ID) utilizing CatBoost classifier (Cb-C) and Lyrebird Optimization Algorithm presented this work (LOA). As typical ID settings, Cb-C excels at handling datasets that are imbalanced. lyrebird’s remarkable capacity imitate sounds its surroundings served as inspiration for LOA, metaheuristic optimization algorithm. WSN-DS dataset, acquired Prince Sultan University Saudi Arabia, used assess suggested method. Among models presented, LOA-Cb-C produces highest accuracy 99.66%; nevertheless, when compared with other methods discussed article, error value 0.34% lowest. Experimental results reveal strategy improves WSN-IoT over existing terms false alarm rate.

Язык: Английский

Intrusion Detection Systems for the Internet of Things Network: Survey on Rare Attacks

Опубликована: Июль 23, 2024

Язык: Английский