Scholarlink

<p>Software Defect Prediction (SDP) plays an important role in enhancing software reliability and reducing the effort required to maintain the product. In this paper, a new framework for SDP is proposed that combines an optimized Variational Autoencoder (VAE) with a Scalable Graph Relearn Attention Network (SGRAN) to address issues of class imbalance, limited labeled data, and overfitting. To aid convergence and data set balancing, a hybrid Particle Swarm Optimization–Chaotic Butterfly Optimization Algorithm (PSO–CBOA) is used in the framework. The optimized VAE is used to learn patterns of latent defects, while SGRAN is used to improve the classification capabilities. The results of the experiment showed that the VAE–CBOA model achieved classification accuracy of 99.07% and an F1-score of 99.01%. In comparison, SGRAN achieved 98.02% accuracy and 98.92% recall, both significantly outperforming traditional methods.</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

Hybrid approach to software defect prediction and classification using variational autoencoder and scalable graph attention network

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Rajesh Kumar Udumu,
  • D.Vasumathi

摘要

Software Defect Prediction (SDP) plays an important role in enhancing software reliability and reducing the effort required to maintain the product. In this paper, a new framework for SDP is proposed that combines an optimized Variational Autoencoder (VAE) with a Scalable Graph Relearn Attention Network (SGRAN) to address issues of class imbalance, limited labeled data, and overfitting. To aid convergence and data set balancing, a hybrid Particle Swarm Optimization–Chaotic Butterfly Optimization Algorithm (PSO–CBOA) is used in the framework. The optimized VAE is used to learn patterns of latent defects, while SGRAN is used to improve the classification capabilities. The results of the experiment showed that the VAE–CBOA model achieved classification accuracy of 99.07% and an F1-score of 99.01%. In comparison, SGRAN achieved 98.02% accuracy and 98.92% recall, both significantly outperforming traditional methods.