Estimating Software Development Efforts Using Random Forest-Based Stacked Ensemble Approach
Abstract
Accurate estimation of software development effort is essential for successful project planning, resource allocation, and cost management, yet it poses significant challenges due to the multifaceted and non-linear relationships among project attributes. Conventional approaches, such as expert judgment, analogy-based estimation, and parametric models like the Constructive Cost Model (COCOMO), often suffer from subjective biases and limited adaptability, leading to unreliable predictions. This study introduces a novel Random Forest-based stacked ensemble model to enhance the precision of software effort estimation. The proposed framework integrates diverse machine learning algorithms, including Random Forest, Support Vector Machines, Gradient Boosting Machines, and Decision Trees, leveraging their complementary strengths. A Random Forest meta- learner aggregates the predictions of these base learners, improving robustness and generalization across varied project contexts. The model was rigorously evaluated on seven benchmark datasets— Albrecht, China, Desharnais, Kemerer, Maxwell, Kitchenham, and Cocomo81—demonstrating superior performance over traditional methods and standalone machine learning models. It achieves significantly lower Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and higher R² scores, indicating better predictive accuracy and explanatory power. By delivering reliable, data- driven effort estimates, this approach supports enhanced project scheduling, budgeting, and resource optimization, offering a scalable and adaptable solution for addressing the complexities of modern software development projects.
Cite this paper
Published in
