A Novel Data Harmonization and Fusion Strategy for Software Cost Estimation Under Limited and Uneven Data Conditions

Estimating software effort became a challenging task in software project management. Most of the research publications focused on identifying the best machine learning algorithms for individual datasets. The effectiveness of these models is often hindered by the limited size and the heterogeneity of available historical project data. To address these limitations, this paper introduced a data fusion and harmonization strategy, we designed sophisticated techniques to harmonize and unify features of nine small size datasets creating an integrated representation for model training. The performance of this strategy was assessed using Random Forest and Gradient Boosting models trained on the resulting fused data. We conducted an empirical evaluation of the performance of the two models. To demonstrate the efficacy of our approach, a comparative study with several machine learning techniques (including individual and ensemble models) trained on each dataset was carried out. It has been proved that data harmonization and fusion approach enhances the prediction performance in terms of mean absolute error (MAE) = 0.854, mean squared error (MSE) = 1.297 and coefficient of determination (R 2 ) = 0.864.