Scholarlink

The expanding intricacy of modern software schemes has made them prone to a varied range of posing severe threats to security vulnerabilities, critical infrastructures and sensitive data. Traditional vulnerability detection approaches often struggle with accuracy and scalability, particularly when managing diverse and large codebases. To deal with these encounters, this research proposes an ensemble framework that incorporates Deep Convolutional Neural Networks (DCNN) with powerful gradient boosting algorithms XGBoost and CatBoost for effective software vulnerability detection. An ensemble approach unites the potencies of both deep learning and gradient boosting to reduce false alarms and enhance detection accuracy. Experimental results reveals that both DCNN along with XGBoost and DCNN along with CatBoost models accomplish high detection operation an accuracy of 87.51%, precision of 87.86%, recall of 79.79%, F1 score of 83.63%, and an AUC PR of 0.9330. In comparison, the DCNN along with XGBoost model accomplishes a slightly better accuracy of 87.51%, precision of 87.93%, recall of 79.72%, F1-score of 83.62%, and an AUC-PR of 0.9332. The results feature the effectiveness of the ensemble approach in recognizing vulnerable code sections, offering a scalable solution for increasing software security in real-world purposes.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CNN-Based Feature Learning with Gradient Boosting for Secure Code Analysis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Himanshu Yadav,
  • Kapil Kumar Nagwanshi

摘要

The expanding intricacy of modern software schemes has made them prone to a varied range of posing severe threats to security vulnerabilities, critical infrastructures and sensitive data. Traditional vulnerability detection approaches often struggle with accuracy and scalability, particularly when managing diverse and large codebases. To deal with these encounters, this research proposes an ensemble framework that incorporates Deep Convolutional Neural Networks (DCNN) with powerful gradient boosting algorithms XGBoost and CatBoost for effective software vulnerability detection. An ensemble approach unites the potencies of both deep learning and gradient boosting to reduce false alarms and enhance detection accuracy. Experimental results reveals that both DCNN along with XGBoost and DCNN along with CatBoost models accomplish high detection operation an accuracy of 87.51%, precision of 87.86%, recall of 79.79%, F1 score of 83.63%, and an AUC PR of 0.9330. In comparison, the DCNN along with XGBoost model accomplishes a slightly better accuracy of 87.51%, precision of 87.93%, recall of 79.72%, F1-score of 83.62%, and an AUC-PR of 0.9332. The results feature the effectiveness of the ensemble approach in recognizing vulnerable code sections, offering a scalable solution for increasing software security in real-world purposes.