Result: Data-Driven Takt Planning: Forecasting Delays and Optimizing Workflow

Traditional progress reporting in Takt-timed construction projects often suffers from reactive monitoring and limited predictive visibility. This study addresses this gap by developing a data-driven early warning system through the implementation of a Multi-Report Analysis Framework (MRAF). Utilizing a longitudinal digital archive transformed from 185 complex progress reports across six distinct BMW infrastructure projects, a robust data engineering pipeline was established. This pipeline features a VBA-driven extraction engine to programmatically harvest weekly work package volumes, followed by a high-performance PySpark framework for distributed cleaning and temporal synchronization. In the analytical phase, descriptive statistics were employed to reconstruct project trajectories and identify trade-specific bottlenecks, utilizing metrics such as the Coefficient of Variation (CV) to quantify flow stability. Subsequently, a Light Gradient Boosting Machine (LGBM) was deployed to train dual-task predictive models for delay occurrence and magnitude. The results demonstrate that the classification model serves as a highly reliable early warning mechanism, achieving an accuracy of 89.88% in Model V2, effectively filtering systemic drift from routine operational noise [cite: 30, 1268, 1302]. While the regression model faced challenges due to the high stochasticity of on-site variables, refinements in Model V2 led to a significant reduction in Mean Absolute Error (MAE) from 36.70% to 20.73%. This research confirms that systematic data refinement and machine learning can effectively transition construction management toward proactive decision support in the era of Construction 4.0.