Scholarlink

<p>Accurate software fault prediction is essential for effective planning, testing, and quality assurance throughout the software development lifecycle. While traditional approaches such as software reliability growth models, fuzzy logic systems, and time series forecasting have been widely used, they often rely on rigid assumptions and lack adaptability to complex and dynamic software environments. In this paper, we propose a weighted voting regression ensemble model aimed at predicting software faults more accurately by using the combined strengths of multiple machine learning algorithms. Seven base regressors are initially evaluated using five-fold cross-validation. The top four models, selected based on root mean squared error, are integrated into a voting ensemble, where each model’s contribution is weighted inversely to its root mean squared error. The proposed model is validated on three real-world software fault datasets derived from large-scale open-source projects. Experimental results demonstrate that the ensemble model outperforms individual learners.</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

A machine learning ensemble for software fault prediction using weighted voting regression

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Umashankar Samal,
  • Chandan Kumar

摘要

Accurate software fault prediction is essential for effective planning, testing, and quality assurance throughout the software development lifecycle. While traditional approaches such as software reliability growth models, fuzzy logic systems, and time series forecasting have been widely used, they often rely on rigid assumptions and lack adaptability to complex and dynamic software environments. In this paper, we propose a weighted voting regression ensemble model aimed at predicting software faults more accurately by using the combined strengths of multiple machine learning algorithms. Seven base regressors are initially evaluated using five-fold cross-validation. The top four models, selected based on root mean squared error, are integrated into a voting ensemble, where each model’s contribution is weighted inversely to its root mean squared error. The proposed model is validated on three real-world software fault datasets derived from large-scale open-source projects. Experimental results demonstrate that the ensemble model outperforms individual learners.