Scholarlink

<p>Software Product Lines (SPLs) aim to reduce development costs and timelines, while improving quality, but their inherent complexity often results in defects and delays. One of the key challenges in SPLs is selecting relevant features from a vast pool of software metrics for defect detection. To address this issue, this study introduces a hybrid approach combining Harris Hawks Optimization (HHO) with stacking-based ensemble learning. The HHO algorithm is further enhanced with the Chaos Optimization Algorithm (COA) to overcome local optima, making it effective for high-dimensional and complex problems. The method is evaluated using the LVAT and NASA repositories, with four datasets from each source. It achieves accuracies of 92.7%, 91.1%, 96.3%, and 98.4% on LTS1, LTM2, LTL3, and LTV4, and 97.91%, 99.01%, 94.21%, and 90.93% on CM1, JM1, KC1, and PC1. The results highlight the proposed method's enhanced efficiency and faster convergence compared to existing approaches.</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

EHHO-EL: a hybrid method for software defect detection in software product lines using extended Harris hawks optimization and ensemble learning

  • 期刊论文

  • 发表日期：

    2025年3月3日

查看全文
  • Mehdi Habibzadeh-khameneh,
  • Akbar Nabiollahi-Najafabadi,
  • Reza Tavoli,
  • Hamid Rastegari

摘要

Software Product Lines (SPLs) aim to reduce development costs and timelines, while improving quality, but their inherent complexity often results in defects and delays. One of the key challenges in SPLs is selecting relevant features from a vast pool of software metrics for defect detection. To address this issue, this study introduces a hybrid approach combining Harris Hawks Optimization (HHO) with stacking-based ensemble learning. The HHO algorithm is further enhanced with the Chaos Optimization Algorithm (COA) to overcome local optima, making it effective for high-dimensional and complex problems. The method is evaluated using the LVAT and NASA repositories, with four datasets from each source. It achieves accuracies of 92.7%, 91.1%, 96.3%, and 98.4% on LTS1, LTM2, LTL3, and LTV4, and 97.91%, 99.01%, 94.21%, and 90.93% on CM1, JM1, KC1, and PC1. The results highlight the proposed method's enhanced efficiency and faster convergence compared to existing approaches.