A Comparative Study to Evaluate the Performance of Selective Models in Software Vulnerability Detection

Software vulnerability detection is one of the essential tasks in cybersecurity; its main target is to identify security defects in software and mitigate them. In this paper, the dataset CVEfixes is investigated and evaluated using various machine learning algorithms: eXtreme Gradient Boosting (XGB), Random Forest (RF), and Naive Bayes (NB), in addition to a deep learning model based on 1D Convolution Neural Network (1D−CNN). A comparison has been conducted between these models, where CNN has outperformed in accuracy but had lower recall and f1-score for the vulnerable samples, indicating weakness in capturing this class. In contrast, XGB, with lower accuracy, had achieved the best f1-score for the vulnerable class.