An Enhanced Network Intrusion Detection System Using ADASYN and Hybrid Residual Block Techniques
Harish G N, H S Annapurna · Engineering Technology & Applied Science Research · 2025
Network security relies on Intrusion Detection Systems (IDS), but current models have problems with feature extraction and inaccurate classification, particularly on unbalanced and small datasets, leading to less effective detection of attack traffic patterns in actual situations. This study suggests an adaptive synthesis method based on an Enhanced Residual Network (ERN) to address these problems to facilitate IDS models in learning from sparse and unbalanced data and increase their detection performance. For optimal distribution of training data and enhanced feature representation, the proposed ERN uses the Inception-ResNet architecture in conjunction with custom sampling modules. Oversampling techniques are used to balance the dataset. The model is trained, tested, and compared to traditional deep learning methods. The proposed model outperforms conventional ones in terms of accuracy, dependability, and feature extraction capacity, as experimentally shown by an intrusion detection accuracy ranging from 89.40% to 91.88%. These results show that the proposed method is a solid choice for tough data settings looking to enhance intrusion detection.