Optimizing Neural Network Architecture for Detecting DDOS Attacks using ANN and XGBoost in Imbalanced Networks
Rissal Efendi · Engineering Technology & Applied Science Research · 2025
The advancement of Internet technology and digital transformation is always followed by increased security concerns in computer networks. Attacks can disrupt services connected to LANs and the Internet, particularly those targeting web-based applications. The most common threat in HTTP is Distributed Denial of Service (DDoS) attacks. Network security is critical to preserve the integrity and availability of services, and it is a critical necessity to have effective methods for detecting and mitigating such attacks to address these risks. Machine learning techniques, particularly ANN and XGBoost, play a key role in enhancing the ability to identify unusual patterns. Despite that, challenges remain in fine-tuning these models for accurate and efficient detection, especially when working with imbalanced data. This study proposes an integrated model that combines ANN with XGBoost to improve detection performance. In the first phase, the ANN architecture is customized to distinguish normal traffic from attacks, while in the second phase, XGBoost is used to refine predictions and improve accuracy. The evaluation results show that the DBSCAN-SMOTE-ANN-XGBoost-PSO model outperforms others, with high accuracy (96.83%), sensitivity (93.23%), and precision (96.13%), demonstrating its effectiveness in detecting DDOS attacks while reducing both false positives and negatives. This integrated approach offers an optimal solution to improve network security and address evolving DDOS attack patterns.