An Intelligent BiGRU-GAP-Dense Framework for Automated Cyber Threat Detection and Proactive Defense

Abstract

The continuous and rapid growth as well as expansion of software systems has significantly increased the chances of hidden vulnerabilities, which is a serious issue in cyber threat detection. The traditional static analysis tools usually fail to detect deeply embedded context-based vulnerabilities. To tackle these challenges, we introduce an integrated deep learning based automated framework cyber threat identification of source code which integrates Bidirectional Gated Recurrent Units (BiGRU), GAP and Dense layers with ReLU function. The main aim of proposed system is to automatically analyze source code snippets and determine whether it is vulnerable or non-vulnerable and if vulnerability is found then it also detects which type of vulnerability either XSS or Command Injection. For that we use Vulnerability Fix Dataset, a publicly available Kaggle dataset containing real-world code in multiple programming languages. To further enrich the training dataset, we extracted additional Python source code from open-source repositories such as Keras and The Algorithms using the PyDriller framework. Preprocessing comprises tokenization, the usage of statistical features, and cleaning of the code to improve quality of input model. The code sequences are then forwarded to created BiGRU- GAP to capture semantic and syntactic patterns. Experiments show that better classification accuracy and better generalization performance can be achieved by the developed network trained with a well-tuned pipeline and on high-performance computing platform. This paper proposed an efficient end-to-end holistic approach for proactive software vulnerability detection by using deep learning.