DEVELOPMENT OF A SOFTWARE VULNERABILITY SCANNER USING SUPPORT VECTOR MACHINES (SVM)

Abstract

Numerous assaults and system failures brought on by the exploitation of vulnerabilities have resulted from the growing reliance on software across industries, underscoring the widespread absence of straightforward detection techniques. Because traditional vulnerability assessment techniques are frequently laborious, resource-intensive, and prone to human error, more accurate and efficient solutions are required. In this study, Support Vector Machines (SVM) were used to construct a software vulnerability scanner. Using the SVM method to analyze software vulnerabilities, designing and implementing the scanner system, and assessing its performance were the specific goals. A mixed-methods approach was used, with the Agile software development process being used. A preprocessed dataset of 100 C/C++ code functions was used to train the SVM model, which was then converted into 527 feature dimensions. A stratified train-test split and five-fold cross-validation were used to assess the model. An unseen test set showed that the developed system, "Identi-fix," performed robustly and had low false-negative rates, with accuracy of 86.2%, precision of 84.8%, recall of 88.0%, and F1-score of 86.4%. The research effectively created an SVM-based vulnerability scanner, which supports the use of machine learning in cybersecurity and improves early software vulnerability identification.