Scholarlink

<p>In the ever-evolving landscape of software defect detection, this study pioneers a robust methodology combining two innovative techniques: the Deep Convolutional RBM Network (DCRBN) and the Kernel-Enhanced Dragonfly PCA approach. The research begins with meticulous data preparation, including data cleaning and min–max normalization, to ensure the integrity and uniformity of the dataset. For feature extraction, this approach harnesses the power of statistical features, such as mean, median, variance, skewness, kurtosis, and various percentiles, providing a comprehensive insight into the underlying data distribution. Additionally, incorporate correlation-based features, enriching the feature set with valuable relationships within the data. To address the issue of high-dimensional data, introduce the Kernel-Enhanced Dragonfly PCA technique, a novel method that effectively reduces dimensionality while preserving essential information, thereby improving the efficiency of subsequent analysis. The core of this defect detection framework lies in the Deep Convolutional RBM Network (DCRBN), a cutting-edge neural network architecture designed to uncover intricate patterns and anomalies within the data. put this methodology to the test using databases sourced from NASA and PROMISE, demonstrating its efficacy in software defect detection with elevated accuracy and reliability. Thus, the performance of the suggested approach can be measured and it can be compared with the existing models. Among them, the suggested approach attains 98.5% of accuracy, 99% of specificity and NPV, 96.5% of sensitivity &amp; precision as well as 96.2% of MCC, F-measure and recall score of the suggested model is about 97%. The FPR of the approach is about 0.002 and FNR of the model is about 0.029 respectively. This research presents a significant step forward in the field, offering a comprehensive and innovative solution for identifying software defects with real-world applications and benefits. However, the limitations in the suggested approach includes, interpretability and computational complexities. In future, investigating the development of lightweight deep learning architectures tailored specifically for defect detection tasks to address computational complexity concerns. Exploring the integration of domain-specific knowledge or contextual information into the model architecture to enhance defect detection accuracy and robustness.</p>

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

Software defect detection via deep convolutional RBM network and kernel enhanced dragonfly PCA approach

  • 期刊论文

  • 发表日期：

    2025年6月16日

查看全文
  • Yang Li,
  • Hong He,
  • Xiaohong Li

摘要

In the ever-evolving landscape of software defect detection, this study pioneers a robust methodology combining two innovative techniques: the Deep Convolutional RBM Network (DCRBN) and the Kernel-Enhanced Dragonfly PCA approach. The research begins with meticulous data preparation, including data cleaning and min–max normalization, to ensure the integrity and uniformity of the dataset. For feature extraction, this approach harnesses the power of statistical features, such as mean, median, variance, skewness, kurtosis, and various percentiles, providing a comprehensive insight into the underlying data distribution. Additionally, incorporate correlation-based features, enriching the feature set with valuable relationships within the data. To address the issue of high-dimensional data, introduce the Kernel-Enhanced Dragonfly PCA technique, a novel method that effectively reduces dimensionality while preserving essential information, thereby improving the efficiency of subsequent analysis. The core of this defect detection framework lies in the Deep Convolutional RBM Network (DCRBN), a cutting-edge neural network architecture designed to uncover intricate patterns and anomalies within the data. put this methodology to the test using databases sourced from NASA and PROMISE, demonstrating its efficacy in software defect detection with elevated accuracy and reliability. Thus, the performance of the suggested approach can be measured and it can be compared with the existing models. Among them, the suggested approach attains 98.5% of accuracy, 99% of specificity and NPV, 96.5% of sensitivity & precision as well as 96.2% of MCC, F-measure and recall score of the suggested model is about 97%. The FPR of the approach is about 0.002 and FNR of the model is about 0.029 respectively. This research presents a significant step forward in the field, offering a comprehensive and innovative solution for identifying software defects with real-world applications and benefits. However, the limitations in the suggested approach includes, interpretability and computational complexities. In future, investigating the development of lightweight deep learning architectures tailored specifically for defect detection tasks to address computational complexity concerns. Exploring the integration of domain-specific knowledge or contextual information into the model architecture to enhance defect detection accuracy and robustness.