A System-Oriented Machine Learning Approach for Planning and Execution of Decisions in Software Project Management

Software project management must make high-stakes decisions under uncertainty in effort estimation, cost control, and execution risk. Although machine learning has enhanced predictive accuracy, several studies employ it only in isolated planning or execution tasks, thereby limiting its usefulness as an end-to-end decision support system. This study describes DeciBoost PM, a single framework that facilitates both planning-layer estimation and execution-layer delay-risk management using a single CatBoost backbone with inherent interpretability. On three heterogeneous public datasets, namely Desharnais to estimate effort, PROMISE to estimate cost, and an Apache JIRA issue-tracking dataset to classify delays and risks, we assess the framework. The same pipeline is used on the datasets, such as preprocessing, feature engineering, leakage-conscious splitting, and equal validation. Standard regression and classification measures are used to measure performance and compare the results with baseline learners. Findings indicate that DeciBoost PM has good and consistent predictive performance with low variance among tasks, thus enhancing better estimation accuracy and delay-risk discrimination. The framework provides transparency in its explanations of SHAP-based and threshold-controlled decision rules that can be directly translated to actionable managerial indicators. In general, DeciBoost PM has captured machine learning as a system-level, practical decision support methodology throughout the software project life cycle.