HYBRID INTELLIGENCE MODELS FOR MULTI-CLASS WEB ATTACK DETECTION AND PREVENTION

Abstract [English]

In today’s digitally driven environment, the frequency and complexity of web-based cyberattacks such as phishing, XSS, and SQL injection have created a pressing need for intelligent and multi-class intrusion detection frameworks. Traditional detection systems often lack adaptability and fail to generalize effectively across diverse attack types. To address this issue, the present study introduces hybrid intelligence models that integrate deep neural architectures including DBM– BiLSTM, GAN with DAE and SAE, and deep residual networks for accurate classification of web attacks.The objective of this research is to develop scalable and high- performing models that can detect multiple classes of attacks with improved accuracy and interpretability. The study utilized synthetically structured datasets, each comprising 1,500 balanced samples across four defined attack classes. Five hybrid models were implemented using TensorFlow and Keras within a Python environment. A standardized preprocessing pipeline involving normalization, label encoding, and data splitting into training (70 percent), validation (15 percent), and testing (15 percent) sets was adopted.Model performance was evaluated using key classification metrics such as accuracy, precision, recall, F1-score, and confusion matrix. Among all the models, the DBM–BiLSTM model demonstrated the highest overall performance, particularly in detecting low- frequency attack types like SQL injection. Based on the findings, the study recommends the use of BiLSTM- based hybrid architectures for real-time intrusion detection. The results highlight the effectiveness of combining temporal learning and deep feature modeling to strengthen cybersecurity systems in complex web environments.