Advancing project success prediction through machine learning-driven models

Although traditional predictive methods support project planning, they often struggle to capture the high-dimensional interactions, nonlinearity, structural complexity, uncertainty and evolving nature of real-world projects. This limitation can reduce predictive accuracy, robustness and probabilistic reliability, thereby constraining managerial responsiveness. In response, this study is motivated by the limited number of field-validated empirical studies that apply advanced ensemble learning techniques, such as calibrated stacking, to model complex relationships among project variables and, in turn, predict project success. Accordingly, the objective of this study is to develop and rigorously evaluate a statistically validated ensemble learning framework for binary project success prediction under realistic, imbalanced project management conditions. The proposed framework integrates feature-selection enhancement, inferential statistical testing, multi-metric evaluation and embedded interpretability diagnostics. The results indicate that calibration-aware stacking with regularized meta-learning delivers robust predictive performance, offering more reliable analytics and stronger decision support for managers operating in complex project environments.