An Improvement of Software Vulnerability and Classification Model using Deep Neural Network Based on Big Data

Abstract

Software vulnerabilities increase the risk of security breaches, potentially causing significant harm to
systems. Automatic classification methods are essential for managing software vulnerabilities and enhancing
system security. This project proposes an improved model named the automatic vulnerability classification
model (information gain based on term frequency-deep neural Network [IGTF-DNN]), which combines
IGTF (term frequency-inverse document frequency [TF-IDF]) and DNN. The TF-IDF method is used to
calculate the frequency and weight of words from vulnerability descriptions and information gain to select
features. An optimal set of feature words is obtained. The DNN then constructs an automatic classifier to
effectively classify vulnerabilities. The model is tested using the National Vulnerability Database of the
United States and shows better performance compared to the K-nearest neighbors’model.