Answers are grounded in this paper and cite their sources.
Web applications are an integral part of digital infrastructure in the modern era and have become increasingly susceptible to advanced cyber threats, thus needing more intelligent and adaptable intrusion detection system. Traditional signature and rule-based (IDS) systems have inherent limitations in detecting various types of web attacks because they are static in nature. This paper presents a multi-class web intrusion detection framework based on deep learning methods by using the WEB-IDS23 dataset, a realistic representation of modern web traffic environments and presents the first holistic multi-class evaluation on this dataset by employing an integrated deep learning and stacking-based ensemble model. The model utilizes Random Forest-based feature selection and combines it with three deep learning models (One-Dimensional Convolutional Neural Network (1D-CNN), Deep Neural Network (DNN) and Gated Recurrent Unit (GRU-RNN)). we then used a stacking-based meta-learning approach to aggregate the predictions. Strong detection performance is demonstrated in experiments. The 1D-CNN yielded best standalone accuracy of 96.43% and F1-score of 96.32%, while stacking ensemble further boosting the classification accuracy to 96.55% and precision to 97.15%. Although the stacking meta-model achieved a marginally lower F1-score compared to its standalone 1D-CNN, it performed with higher overall accuracy. This phenomenon is due to the imbalanced class distribution of the WEB-IDS23 dataset, where optimizing majority-class classification raised overall accuracy but lowered recall and precision on some minority classes. By combining these methods with feature selection, deep learning and stacking-based ensemble learning, the final result achieves higher accuracy in detection task as well as robust detection capability and scalability. The proposed system is an efficient approach for the multi-class web intrusion detection system
No takes yet. Share an insight, caveat, or question.
Ziyad et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: