In Pre-Engineered Building (PEB) production, forecasting production timelines accurately is crucial, especially as tasks like cutting, welding, and painting vary depending on structural complexity and job size. Many conventional scheduling methods fall short when dealing with the dynamic, multi-stage nature of fabrication workflows. This study introduces a profitability-aware machine learning–based approach that predicts both overall and stage level fabrication durations using Random Forest regression. The model is trained on part-level production data, with features including project tonnage, number of parts, and statistical descriptors of task durations. To improve learning performance, projects are grouped into six distinct complexity classes based on fabrication characteristics. A profitability-oriented evaluation is also proposed, which labels prediction outcomes as Profitable, Acceptable, or Excess depending on how closely they align with business targets. The model is tested on real data from 34 completed PEB projects, showing clear improvements over conventional estimation methods. Results demonstrate that stage-level predictions outperform start-date-based forecasts, ensuring profitability-aligned outcomes even in high-complexity projects. The proposed framework helps bridge the gap between technical forecasting accuracy and real-world production goals, offering a practical solution for smarter planning in steel fabrication environments.
