Applying Machine Learning to Estimate the Effort and Duration of Individual Tasks in Software Projects

CONTEXT: Software estimation is a vital yet challenging project management activity. Various methods, from empirical to algorithmic, have been developed fit different development contexts, plan-driven agile. Recently, machine learning techniques shown potential in this realm but are still underexplored, especially for individual task estimation. OBJECTIVES: We investigate the use of predicting effort and duration software projects assess their applicability effectiveness production environments, identify best-performing algorithms, pinpoint key input variables (features) predictions. METHOD: conducted experiments with datasets various sizes structures exported three tools used by partner companies. For each dataset, we trained regression models tasks using eight algorithms. The were validated k-fold cross-validation evaluated several metrics. RESULTS: Ensemble algorithms like Random Forest, Extra Trees Regressor, XGBoost consistently outperformed non-ensemble ones across datasets. However, accuracy feature importance varied significantly datasets, Mean Magnitude Relative Error (MMRE) ranging 0.11 9.45 target variables. Nevertheless, even worst-performing estimates aggregated level showed good accuracy, MMRE = 0.23. CONCLUSIONS: Machine ensemble ones, seem be viable option estimating projects. quality relevant features may depend largely on characteristics available underlying when poor, at present due error compensation.