An Intelligent Hybrid Deep Learning Model for Proactive Cyber Threat Detection and Defense

Abstract

The research presents a hybrid AI based deep learning model that adds Bidirectional GRU (BiGRU) and Global Average Pooling (GAP) to detect software vulnerabilities. The main goal of this model is to find security issues in software source code, like SQLi and XSS attacks. The BiGRU helps in understanding the code by analysing it in both past and future directions, which improves the detection of patterns and relationships in the code. The GAP layer helps to reduce the size of the dense vector and prevents the model from over fitting. Overall, this approach improves the accuracy and efficiency of cyber threat detection. The researchers assess the proposed model using real-world vulnerability datasets which they collected from Kaggle and which contain labelled code samples of both vulnerable and non-vulnerable instances. The experimental results show that the proposed model performs well from all baseline models which include RNN, LSTM, GRU, and BiGRU because it achieves an accuracy rate of 97.5% and a precision rate of 98% and a recall rate of 97% and an F1-score of 97.5%. The framework provides perfect and scalable solution for early vulnerability detection which enables integration into DevSecOps pipelines while it improves software security through proactive identification of potential threats.