Proactive DDoS attacks detection using deep learning techniques
Ravi Kumar Gurugubelli, Krishna Naick Bhukya, Ramesh Choppa · IET conference proceedings. · 2025
In this study, present a comprehensive analysis of detecting Distributed Denial of Service (DDoS) attacks using advanced deep learning models, including a Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Long Short - Term Memory (LSTM) network. Leveraging the CICDDoS2019 dataset, we evaluate the performance of these models across several metrics, such as accuracy, precision, recall, F1-score, log-loss, and ROC-AUC, to determine their efficacy in distinguishing DDoS attack traffic from normal traffic. The DNN model achieves an impressive accur acy of 98.31% and near-perfect precision and recall values, highlighting its robustness in capturing the characteristics of DDoS attacks. The CNN model, while effective, shows a slightly lower accuracy of 97.27%, indicating its limitations in processing c omplex sequential dependencies. The LSTM model, designed to learn from temporal patterns, demonstrates a solid performance with an accuracy of 96.78% and high recall, making it suitable for sequence-based detection scenarios. Furthermore, our analysis of log-loss values reveals that the LSTM model achieves the lowest value, indicating strong model confidence, while the ROC - AUC values emphasize the exceptional discriminative capabilities of all models, with the DNN model performing the best. These results underline the potential of deep learning techniques in enhancing DDoS attack detection, with each model offering distinct strengths depending on the data characteristics and detection requirements.