Publication

Abstract : This survey examines the use of AI and machine learning in software vulnerability detection. As software becomes more advanced and complicated over the days cyber-attacks get more sophisticated and advanced, so this has become a problem, this survey examines deep learning methods like CNN, RNN & BiLSTMs, transformers like BERT and all its variations, it also examines graph-based methods like GGNNs and HAGNNs, and conventional ML methods. Key datasets include SARD, NVD, Debian Linus repositories, and Python-specific datasets have been used and examined for their advantages and disadvantages. By comparing performance metrics including accuracy, F1-score, recall, precision, and ROC, reliability is evaluated. To promote secure engineering methods, the study indicates future prospects for multimodal learning, cross-language transfer, federated learning, and integration with software pipelines. It also addresses long-standing challenges such as language reliance, dataset bias, scalability, and explainability.