The forecast of effort needed to complete software development is essential to control costs, schedules, and allocation of resources in the current projects. However, initial estimates are often inaccurate because of the multifaceted nature of project data. This is the kind of data that is usually inconsistent, non-linear relationships, and the effect of both human and technical aspects, which classical models can hardly address entirely account for. The machine proposed in this paper is a novel, two-part machine method of learning that is used to approximate software development effort and spot projects likely to go over budget limits. The framework takes a blended strategy: a stacked ensemble regressor, a combination of Random Forest and Gradient Enhancing algorithms, to generate accurate effort predictions, and a high-risk project identification layer to secondary classification layer of budget overruns. Experiment on the Desharnais and NASA93 datasets showed interesting outcomes, as the model obtained a Mean Magnitude of Relative Error (MMRE) of 0.138 and a classification accuracy of 91.5%. Additional discussion of error rates means that the model provides reliable risk estimates and without producing too many false alarms.

Keywords- Software Effort Estimation, Machine Learning, Stacked Ensemble, Risk Classification, Random Forest, Gradient Boosting, MMRE.