Material Forecasting in PEB Fabrication Using Machine Learning Techniques

Abstract

This paper presents a machine learning-based predictive model to enhance material forecasting and resource optimization in Pre-Engineered Building (PEB) production. Random Forest Regression, along with GridSearchCV for hyperparameter optimization and training, is utilized in the proposed model. A dataset of 70+ actual-time project executions with careful filtering based on stringent disaggregation requirements, including fabrication completeness, critical components of PEB, and quantity limitations, is trained. Projects are probabilistically differentiated into P1 (AND) and P2 (OR) types and also differentiated into fabrication combinations (C1, C2, C3) based on structural complexity and member contribution. A subset of 47 highly used and influential consumables is chosen from 100 plus consumables commonly used to enhance model performance. The proposed predictive model identifies a profitability percentage of 80% in all the grouped datasets, thereby asserting its feasibility for real-world use. The proposed methodology facilitates proper material planning, minimizes waste in fabrication, and aids in strategic decision-making for medium-to-large-scale industrial PEB projects, thereby driving sustainability and increasing operational efficiency.