Puguh Jayadi, Didik Dwi Prasetya, Triyanna Widiyaningtyas
Estimating effort in software development is a crucial aspect that affects the overall success of the project. Traditional methods such as Use Case Point (UCP) are often less accurate, and more sophisticated Machine Learning (ML) models tend to be black-boxed, making interpretation difficult. This study aims to use and understand the software effort estimation model using Explainable AI (XAI) techniques. By utilizing the public UCP dataset, which consists of 71 projects, the Linear Regression and Decision Tree models were trained and evaluated. The results showed that the Decision Tree model provided significantly better predictive performance, with MAE of 198.35 hours of people and NMAE of 0.0947, compared to the Linear Regression model, which had MAE of 671.23 hours of people and NMAE of 0.3205, and the baseline method, which had AUCP of 20. Furthermore, XAI techniques, including the built-in feature importance of Decision Trees and SHapley Additive exPlanations (SHAP) analysis, were applied to interpret the Decision Tree model. SHAP results consistently identified Unadjusted Actor Weight (UAW) and Unadjusted Use Case Weight (UUCW) as features with the most significant contribution to effort estimation, providing transparent insights into the key driving factors. This study demonstrates that the combination of a relatively simple ML model with XAI techniques can significantly improve understanding and confidence in software effort estimation models. © 2025 IEEE.
Universitas Negeri Malang, Department of Electrical Engineering and Informatics, Malang, Indonesia; Universitas Pgri Madiun, Department of Informatics, Madiun, Indonesia
