Scholarlink

<p>Scrum, an agile methodology emphasizing iterative development and adaptive planning, facilitates high-quality software delivery. However, achieving accurate effort estimation (measured in story points) for Scrum projects is often challenged by factors such as shifting requirements, varying task complexities, and inherent biases in traditional estimation techniques. To address these limitations, this study proposes and rigorously evaluates an ensemble Machine Learning (ML) model, integrating Ridge Regression (RR), Extra Trees (ET), and Multi-Layer Perceptron (MLP). The model’s efficacy was confirmed through a multi-faceted evaluation. Firstly, on a public dataset of 140 user stories, it achieved strong performance Mean Absolute Error (MAE) of 0.68, Mean Square Error (MSE) of 0.74, Root Mean Square Error (RMSE) of 0.86. This improvement was demonstrated to be statistically significant (p &lt; 0.05) via the Wilcoxon signed-rank test. Secondly, to address the critical research gap of generalizability, the model’s robustness was validated on a secondary, real-world dataset of 159 user stories, where it maintained its superior performance (MAE=0.70, MSE=0.90, RMSE=0.95). Finally, a detailed error analysis provides actionable insights for practitioners, linking the model’s performance to the agile best practice of decomposing large user stories. The findings validate that the proposed ensemble model offers a statistically robust, generalizable, and practically applicable data-driven solution for enhancing sprint planning and resource allocation in agile project management.</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

Effort estimation in scrum using AI

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Maria Rasheed,
  • Iman Fatima,
  • Darain Fatima,
  • Muhammad Hamid,
  • Fahima Hajjej

摘要

Scrum, an agile methodology emphasizing iterative development and adaptive planning, facilitates high-quality software delivery. However, achieving accurate effort estimation (measured in story points) for Scrum projects is often challenged by factors such as shifting requirements, varying task complexities, and inherent biases in traditional estimation techniques. To address these limitations, this study proposes and rigorously evaluates an ensemble Machine Learning (ML) model, integrating Ridge Regression (RR), Extra Trees (ET), and Multi-Layer Perceptron (MLP). The model’s efficacy was confirmed through a multi-faceted evaluation. Firstly, on a public dataset of 140 user stories, it achieved strong performance Mean Absolute Error (MAE) of 0.68, Mean Square Error (MSE) of 0.74, Root Mean Square Error (RMSE) of 0.86. This improvement was demonstrated to be statistically significant (p < 0.05) via the Wilcoxon signed-rank test. Secondly, to address the critical research gap of generalizability, the model’s robustness was validated on a secondary, real-world dataset of 159 user stories, where it maintained its superior performance (MAE=0.70, MSE=0.90, RMSE=0.95). Finally, a detailed error analysis provides actionable insights for practitioners, linking the model’s performance to the agile best practice of decomposing large user stories. The findings validate that the proposed ensemble model offers a statistically robust, generalizable, and practically applicable data-driven solution for enhancing sprint planning and resource allocation in agile project management.