Learning-Based Multi-Criteria Decision Making Model for Sawmill Location Problems
A new ML framework uses Random Forest and SHAP to identify the most influential factors for industrial location.
A team of researchers has published a novel AI framework designed to solve complex industrial location problems. Their paper, "Learning-Based Multi-Criteria Decision Making Model for Sawmill Location Problems," introduces an LB-MCDM model that fuses machine learning with Geographic Information System (GIS)-based spatial analysis. The goal is to provide a replicable, data-driven method for strategically placing sawmills to maximize efficiency, profitability, and sustainability in timber supply chains. The researchers demonstrated the model's utility through a detailed case study in Mississippi.
The team tested five prominent ML algorithms—Random Forest Classifier, Support Vector Classifier, XGBoost Classifier, Logistic Regression, and K-Nearest Neighbors—to assess site suitability. The Random Forest Classifier emerged as the top performer. To ensure interpretability, they used SHAP (SHapley Additive exPlanations) analysis to identify the most critical decision criteria. This revealed that the composite Supply-Demand Ratio, reflecting local market competition, was the most influential factor, followed by proximity to roads, rail lines, and urban areas. The final validation showed that approximately 10-11% of Mississippi's total land area is classified as highly suitable for sawmill development based on their model's analysis.
- The framework combines five ML algorithms with GIS and MCDM for a data-driven site selection process.
- SHAP analysis identified the Supply-Demand Ratio as the #1 factor, beating infrastructure proximity.
- The model classified 10-11% of Mississippi as 'highly suitable,' providing a concrete, actionable output for planners.
Why It Matters
This demonstrates AI's move into tangible industrial optimization, offering a blueprint for data-driven logistics and supply chain planning beyond forestry.