The Zoneout Regularized Gated Recurrent Unit Algorithm for Network Intrusion Detection with Class Imbalance Mitigation
K. Mala, H S Annapurna · Engineering Technology & Applied Science Research · 2025
This study used a Zoneout-Regularized Gated Recurrent Unit (ZR-GRU) to enhance the effectiveness of Network Intrusion Detection Systems (NIDSs) by maintaining long-term temporal patterns and reducing the risk of overfitting. In contrast to traditional models, ZR-GRU incorporates zoneout regularization to improve generalization over diverse network traffic patterns. To address the prevalent issue of class imbalance in network security datasets, the model integrates the Synthetic Minority Oversampling Technique (SMOTE) for oversampling minority classes and NearMiss for undersampling majority classes, promoting balanced class representation. The model was evaluated on three widely used benchmark datasets, UNSW-NB15, CICIDS 2018, and CIC-DDoS 2019, chosen due to their realistic network traffic characteristics and the diverse range of contemporary attack types. ZR-GRU achieved high accuracy rates of 99.91%, 99.92%, and 99.14% on these datasets, outperforming traditional architectures. The findings highlight the strength, flexibility, and effectiveness of the model for real-time and adaptive intrusion detection in diverse network settings.