The goal of the current study is to conduct a comparative analysis of neural network architectures based on vulnerabilities identified by Open Worldwide Application Security Project in a web application context. This study comes from a descriptive and quasi-experimental model and is real data based empirical research. Moreover, in this study we identify the study as applied as the research aims to establish practical knowledge in a subject area. Our dataset is historical (post event), and the research design used is descriptive-correlational. We develop and analyze four different types of neural networks: Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) and Hybrid Neural Networks (Hybrid NNs) in our experiment. To accurately predict, identify and block cyberattacks, the structures were trained on past internet usage data. The research design involved the design of the web page, data collection for cyberattacks and normal operational data, data cleaning, design of the neural network, training of the network, integration of the network inside the web page, blocking mechanism for malicious requests, and performance evaluation of the neural networks. The models resulted in precision, recall and F1 values, together with an area under the ROC curve of 0.98, that reflects their effectiveness in appropriately segmenting related data. Furthermore, none of the models showed a risk of overfitting since they had approximately identical accuracy levels for both training and validation set, with no significant discrepancies noted. After training using FLASK API the models were also added in web page. After running high-intensity OWASP ZAP attacks it was observed that this program can spot the attacks efficiently and block an end user by entering bad information. In addition, the predictive, detecting, and blocking of cyber-attacks by three neural networks at the same rate (NN, RNN, and Hybrid) was 90% and the entire attack frequency was observed. Index Terms: Web application security, artificial neural networks, convolutional neural networks, recurrent neural networks, hybrid neural networks.
