Regression Model Exploration in Software Effort Estimation Using the Use Case Point Method

Abstract

Software development effort estimation is a crucial aspect of project management as it directly affects scheduling, resource allocation, and cost control. The Use Case Point (UCP) method is widely used for early-stage estimation; However, its traditional approach has several limitations, particularly related to subjective assessments and the tendency toward overestimation or underestimation. This study aims to explore and compare the performance of various regression models in improving the accuracy of UCP-based effort estimation. The dataset consists of 71 completed software projects, using UAW, UUCW, TCF, and ECF as predictor variables, and actual effort as the target variable. The evaluated models include Polynomial Regression, Decision Tree, Random Forest, Gradient Boosting, and Ridge Regression. Model performance was assessed using Mean Absolute Error (MAE), Mean Balanced Relative Error (MBRE), and Mean Inverted Balanced Relative Error (MIBRE) with an 80:20 train–test split. The experimental results indicate that the optimized Random Forest model achieves the best balance between training accuracy and generalization ability on unseen data (test MAE of 11.38), significantly outperforming the traditional UCP calculation method (MAE of 90.33). These findings suggest that non-linear regression approaches, particularly ensemble-based methods, can substantially enhance the reliability of software effort estimation compared to the conventional UCP method.