Advancing Environmental Informatics: A Review Of Hybrid Ensemble Models For Software Effort Estimation In Eco-Systems

Abstract

redicting the amount of work needed to create or maintain software applications is known as software effort estimation, and it is an essential task in software project management. Software projects must be completed on schedule and within budget, and efficient planning and staffing are made possible by accurate estimations. This paper examines the application of machine learning techniques to enhance software effort estimation using actual datasets. We used five publicly accessible datasets: NASA93, COCOMO, Maxwell, Desharnais, and ISBSG. Handling missing values, transforming categorical features, and dividing the data into train-test sets were all examples of pre-processing. Decision trees, Random Forest, Gradient Boosting, and linear regression were the four machine learning regression techniques that were assessed. To choose a pertinent subset of traits and lower dimensionality, correlation-based feature selection was also used. In order to assess prediction accuracy, the comparison study concentrated on two important metrics: R2 and root mean squared error (RMSE). The findings show that Random Forest and linear regression models outperform other methods by a significant margin when using correlation to choose features for this effort estimate job. The datasets with the greatest R2 values were NASA93, COCOMO, Maxwell, and Desharnais. With the lowest RMSE, the Desharnais dataset appears to be very accurate. According to the results, machine learning models for estimating development effort can be enhanced by correlation-based feature selection. The advantages of Random Forest and linear regression models allow them to be used to create trustworthy estimate tools. The data obtained from this comparative analysis provides a solid foundation for further investigation.