From Critical Path to Cost Award: An Explainable XGBoost Model for US Construction Delay Court Decisions
Ahmed Y. El Sayed, Ola Diaa El Monayeri, Khalid S. Al-Gahtani, Walid Mostafa El Nemr, Mohamed Kamel Elshaarawy, Hazem Khaled Aboulata Mohamed, Mahmoud ElbialyAbstract
Forecasting monetary outcomes in construction delay litigation remains difficult because cost awards are shaped by interacting factual, evidentiary, contractual, and forum-specific considerations rather than by incurred cost alone. This study develops interpretable machine learning models to estimate cost awards from published US construction delay decisions, with separate federal court and state court models to avoid pooling materially different adjudicative environments. The case database was assembled from Google Scholar case law searches using “delay” AND “critical path,” followed by eligibility screening, deduplication, and evidence-traceable coding. The coding framework was first piloted using 33 candidate factors and then reduced to 29 retained factors after excluding sparsely and inconsistently reported variables. Predictors were harmonized using controlled vocabularies, missing categorical information was coded as an explicit Not Specified category rather than imputed by mode, and monetary awards were modeled on a log-transformed scale. Extreme gradient boosting (XGBoost) was selected because it can model nonlinear interactions among heterogeneous categorical predictors while remaining interpretable through global feature importance diagnostics. Using an 80/20 train–test evaluation, the federal model achieved test