This study proposes a hybrid linear–nonlinear regression framework for early-stage prediction of construction schedule delay by integrating environmental indicators with Earned Value Management (EVM) metrics during the first 20% of project duration.
Environmental variables (temperature, rainfall, humidity, wind speed and mSPI) were combined with EVM indicators (PV, EV, AC, SPI and CPI) to predict the schedule delay ratio (SDR). Seventy-eight weekly observations from three building projects were analysed. A multiple linear regression (MLR) model was integrated with nonlinear models (Decision Tree and XGBoost) using an optimised fixed-weight fusion strategy. Time-ordered validation, cross-validation and sensitivity analysis were employed to ensure robustness.
The hybrid model achieved a 20–50% reduction in RMSE compared with standalone models and demonstrated improved residual stability and lower prediction variance. Rainfall and mSPI were identified as dominant early-stage predictors, while SPI and CPI provided complementary managerial insight. Sensitivity analysis confirmed the nonlinear amplification of adverse climatic effects.
The study is based on three region-specific projects and a limited dataset; results should be interpreted as proof-of-concept evidence.
The framework enables early warning of delay risk, facilitating proactive resource allocation and schedule intervention.
Improved delay prediction supports timely project delivery, reducing cost overruns, resource wastage and environmental impacts, while enhancing workforce stability and stakeholder confidence.
This study introduces a parsimonious fixed-weight linear–nonlinear fusion framework tailored for small-sample, early-stage construction data. By jointly modelling environmental variability and EVM indicators, the research enhances predictive accuracy while maintaining interpretability, supporting proactive schedule risk management.
