Enhancing Software Effort Estimation: A Comparative Analysis of Machine Learning Models with Correlation-Based Feature Selection

Muhammad Abid, Department of Mathematics, North Carolina State University, Raleigh, 27606, NC, United StatesFollow
Momina Liaqat Ali, Department of Computer Science, Middle Tennessee State University, 1301 E Main St. Murfreesboro, TN, 37132, United StatesFollow

ORCID

Muhammad Abid: https://orcid.org/0009-0006-8619-6666

Momina Liaqat Ali: https://orcid.org/0000-0003-0180-6965

Keywords

Software effort estimation, Machine learning, Feature selection, Predictive modeling, Data analysis

Article Type

Original Article

Abstract

Software effort estimation is a critical process in the management of software projects, essential for predicting the resources and time required for application development. Accurate estimation is paramount for delivering projects on time and computational resources. This paper investigates techniques that utilize machine learning models to ameliorate software effort estimation. We employ four popular machine learning models including the Linear Regression, Gradient Boosting, Random Forest, and Decision Tree and evaluate their performance across five diverse climate-related datasets: Antarctica Hotpoints/Climate Change NASA, Crop Production and Climate Change, Global Warming on Earth (Global Temp), Climate Change, and Climate Treasure (Marine). A key focus is the influence of correlation-based feature selection on model reliability and predictive accuracy. For evaluation, R-squared (R²) and Root Mean Squared Error (RMSE) are employed. The outcomes delineate that correlation-based feature selection consistently and substantially optimizes the performance of all models across these datasets. Linear Regression demonstrated exceptional performance on datasets with predominantly linear relationships, achieving a near-perfect R² of 0.999 and an RMSE of 0.0000000007 on Crop Production, and an R² of 0.98 and RMSE of 6.3 on Climate Treasure (Marine). Ensemble models, Gradient Boosting and Random Forest, consistently delivered strong results on more complex datasets, with Gradient Boosting achieving an R² of 0.94 and RMSE of 0.12 on Antarctica Hotpoints, and Random Forest an R² of 0.70 and RMSE of 22900.00 on Global Warming. This juxtaposed analysis provides a solid framework for future research, enabling project managers to build more fact-based effort predictions.

How to Cite

Abid, Muhammad and Ali, Momina Liaqat (2025) "Enhancing Software Effort Estimation: A Comparative Analysis of Machine Learning Models with Correlation-Based Feature Selection," Sustainable Machine Intelligence Journal: Vol. 12: Iss. 1, Article 1.
DOI: 10.61356/SMIJ.2025.12564
Available at: https://smij.sciencesforce.com/journal/vol12/iss1/1