Scholarlink

<p>Accurately predicting software defect density (SDD) is crucial during the early stages of software development, as it ensures high-quality software delivery and reduces future maintenance costs. SDD serves as a vital metric for evaluating software quality; however, its prediction is often hindered by the presence of irrelevant and redundant features. Since feature selection is an NP-hard problem, identifying the most relevant attributes significantly enhances predictive performance. To address this, we propose a novel Binary Artificial Hummingbird Algorithm (BAHA) that leverages both S-shaped and V-shaped transfer functions to effectively convert continuous values into binary representations suitable for feature selection tasks. The proposed BAHA is applied to 30 datasets from the PROMISE repository to optimize feature subsets and improve the performance of machine learning regressors. Experimental results demonstrate that BAHA significantly outperforms traditional feature selection methods, achieving substantial improvements in predictive accuracy. The BAHA-XGBoost model achieved the lowest Mean Absolute Error (MAE) of 0.24152, outperforming AHA-XGB by 95.98% and XGB without feature selection by 22.7%. Additional metrics such as MSE (32.13), RMSE (2.10), and R<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(^2\)</EquationSource> </InlineEquation> (0.95) confirm the robustness of the approach. Statistical validation using the Friedman test ranks BAHA as the top-performing optimizer among various metaheuristic techniques, highlighting its superior capability in selecting informative features for SDD prediction.</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

BAHA optimization based feature selection approach for software defect density modeling using S and V transfer function

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Jasmeet Kaur,
  • Arvinder Kaur,
  • Kamaldeep Kaur

摘要

Accurately predicting software defect density (SDD) is crucial during the early stages of software development, as it ensures high-quality software delivery and reduces future maintenance costs. SDD serves as a vital metric for evaluating software quality; however, its prediction is often hindered by the presence of irrelevant and redundant features. Since feature selection is an NP-hard problem, identifying the most relevant attributes significantly enhances predictive performance. To address this, we propose a novel Binary Artificial Hummingbird Algorithm (BAHA) that leverages both S-shaped and V-shaped transfer functions to effectively convert continuous values into binary representations suitable for feature selection tasks. The proposed BAHA is applied to 30 datasets from the PROMISE repository to optimize feature subsets and improve the performance of machine learning regressors. Experimental results demonstrate that BAHA significantly outperforms traditional feature selection methods, achieving substantial improvements in predictive accuracy. The BAHA-XGBoost model achieved the lowest Mean Absolute Error (MAE) of 0.24152, outperforming AHA-XGB by 95.98% and XGB without feature selection by 22.7%. Additional metrics such as MSE (32.13), RMSE (2.10), and R \(^2\) (0.95) confirm the robustness of the approach. Statistical validation using the Friedman test ranks BAHA as the top-performing optimizer among various metaheuristic techniques, highlighting its superior capability in selecting informative features for SDD prediction.