Meta-heuristic-based hybrid deep learning model for vulnerability detection and prevention in software system

Abstract

Software vulnerabilities are flaws that may be exploited to cause loss or harm. Various automated machine-learning techniques have been developed in preceding studies to detect software vulnerabilities. This work tries to develop a technique for securing the software on the basis of their vulnerabilities that are already known, by developing a hybrid deep learning model to detect those vulnerabilities. Moreover, certain countermeasures are suggested based on the types of vulnerability to prevent the attack further. For different software projects taken as the dataset, feature fusion is done by utilizing canonical correlation analysis together with Deep Residual Network (DRN). A hybrid deep learning technique trained using AdamW-Rat Swarm Optimizer (AdamW-RSO) is designed to detect software vulnerability. Hybrid deep learning makes use of the Deep Belief Network (DBN) and Generative Adversarial Network (GAN). For every vulnerability, its location of occurrence within the software development procedures and techniques of alleviation via implementation level or design level activities are described. Thus, it helps in understanding the appearance of vulnerabilities, suggesting the use of various countermeasures during the initial phases of software design, and therefore, assures software security. Evaluating the performance of vulnerability detection by the proposed technique regarding recall, precision, and f-measure, it is found to be more effective than the existing methods.