Near-Term Performance Early Warning in Building Projects Using XGBoost and Earned Value Metrics

Abstract

Machine learning can contribute to more proactive, data-driven construction project management. However, most research efforts have focused on predicting final project outcomes, which may miss short-term variations that reveal emerging risks, thereby limiting the utility of their solutions as practical predictive tools for on-site project control. This paper develops and evaluates an early-warning machine learning model based on extreme gradient boosting (XGBoost) to classify next-month performance of building construction projects (Good, At-Risk, or Poor) using earned value management (EVM) and earned schedule (ES) metrics. Monthly EVM/ES records from 17 building projects were used along with schedule and cost performance indices to build the dataset. Three acceleration features were proposed to capture the trend in project performance. The model was trained on a time-aware experimental design using expanding-window cross-validation, light XGBoost tuning, probability calibration, and class-specific decision thresholds, and evaluated on a chronologically held-out test set. The model yielded a Macro- F 1 score of 0.760 on the test set, with a recall of 91% for “Poor” and 82% for “Good” classes, and a conservative behavior for “At-Risk” cases. Permutation importance and Shapley additive explanations (SHAP) analyses indicated that schedule indices are the main drivers of predicted performance outcomes, with cost and acceleration indicators providing complementary signals. In contrast to existing ML-based EVM studies that focus on final-outcome prediction, this study proposes a near-term multiclass early-warning approach that relies solely on standard EVM/ES metrics and achieves acceptable predictive performance using a relatively small dataset, providing guidance for designing feasible early-warning solutions under limited data availability and aligned with field managers’ information needs. The model can be implemented on-site using monthly control data to provide timely alerts on next-month project performance and support more proactive, data-driven project control.