Software Vulnerability Assessment and Classification Using Machine Learning, Deep Learning and Feature Selection Techniques

Abstract

The detection of software defects is a critical technique for improving software quality and optimizing testing resources. This study presents a novel approach to software vulnerability assessment and classification using Recurrent Neural Networks (RNNs) enhanced by feature selection techniques. The proposed methodology integrates data preprocessing, dynamic analysis methods, and vector space model (VSM) generation, leveraging techniques such as TF-IDF and relational feature extraction to normalize and balance datasets. Computational experiments were conducted using various real-world and synthetic datasets, comparing the proposed RNN framework to traditional machine learning models, including Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Adaboost. The RNN model, optimized with activation functions such as ReLU, Sigmoid, and Tanh, demonstrated superior performance, achieving a classification accuracy of 97.5% with ReLU and outperforming other models in precision (97.6%), recall (97.9%), and F-measure metrics. These results highlight the robustness and effectiveness of the proposed framework in detecting vulnerabilities and mitigating software defects. This research underscores the potential of deep learning-based approaches in enhancing software reliability and security.