Learning-Based Multi-Criteria Decision Making Model for Sawmill Location Problems

Mahid Ahmed; Ali Dogru; Chaoyang Zhang; Chao Meng

cs.LGApr 5, 2026

Learning-Based Multi-Criteria Decision Making Model for Sawmill Location Problems

Authors: Mahid Ahmed, Ali Dogru, Chaoyang Zhang, Chao Meng

Organizations: School of Computing Science & Computer Engineering, University of Southern Mississippi · School of Management, University of Southern Mississippi · School of Marketing, University of Southern Mississippi

Abstract

Strategically locating a sawmill is vital for enhancing the efficiency, profitability, and sustainability of timber supply chains. Our study proposes a Learning-Based Multi-Criteria Decision-Making (LB-MCDM) framework that integrates machine learning (ML) with GIS-based spatial location analysis via MCDM. The proposed framework provides a data-driven, unbiased, and replicable approach to assessing site suitability. We demonstrate the utility of the proposed model through a case study in Mississippi (MS). We apply five ML algorithms (Random Forest Classifier, Support Vector Classifier, XGBoost Classifier, Logistic Regression, and K-Nearest Neighbors Classifier) to identify the most suitable sawmill locations in Mississippi. Among these models, the Random Forest Classifier achieved the highest performance. We use the SHAP (SHapley Additive exPlanations) technique to determine the relative importance of each criterion, revealing the Supply-Demand Ratio, a composite feature that reflects local market competition dynamics, as the most influential factor, followed by Road, Rail Line and Urban Area Distance. The validation of suitability maps generated by our LB-MCDM model suggests that 10-11% of the MS landscape is highly suitable for sawmill location.

Figures & tables

Data	Source	Measurement	Purpose
National Land Cover Data	US Census Bureau	Highly suitable: Developed areas located on barren land with open space and low intensity, Suitable: Forest land and grassland, Somewhat suitable: Developed areas located on shrubland and cultivated land with medium and high intensity, Not suitable: Water space, and wetlands.	To distinguish candidate locations based on their land suitability and potential for sustainable development.
Transportation Data	USGS National Transportation Dataset	Highly suitable: 0m $<$ Value $\leq$ 500m, Suitable: 500m $<$ Value $\leq$ 1000m, Somewhat suitable: 1000m $<$ Value $\leq$ 2000m, Not suitable: Value $>$ 2000m. Source: [ 3 ]	To measure proximity to major roads and rail lines in order to improve transportation access and reduce logistical costs.
Slope Data	USGS open topography	Highly suitable: $0\leq\text{Slope}\leq 7\%$ , Suitable: $8\leq\text{Slope}\leq 13\%$ , Somewhat suitable: $14\leq\text{Slope}\leq 20\%$ , Not suitable: $21\leq\text{Slope}\leq 90\%$ . Source: [ 4 ]	To determine the slope of potential sites to ensure ease of access and support more efficient transportation and construction.
Urban Data	US Census Bureau	Highly suitable: Value $<10{,}000m$ , Suitable: $10{,}000m\leq\text{Value}<20{,}000m$ , Somewhat suitable: $20{,}000m\leq\text{Value}<50{,}000m$ , Not suitable: $\text{Value}\geq 50{,}000m$	To ensure better access to utilities and services such as electricity, gas, water, communication, banking, and healthcare facilities.
Labor Data	U.S. Bureau of Labor Statistics	County-level unemployment rate for labor whose age is greater than 16.	To assess the availability of labor in the area to support sawmill operations.
Timber Demand Data	Forisk Mill Capacity Database	The data provides detailed information on sawmills’ timber processing capacities and demands in tons.	To align site selection with existing timber demand and operational capacity constraints.

Table 1: Summary of Datasets

Panel of Figure 2 — Figure 2: Comparison of Timber Supply, Timber Demand, and Supply Demand Ratio in MS

Figure 3: Correlation Matrix among the 9 input features

Feature	VIF	TOL
Road Distance	1.0762	0.9292
Rail Line Distance	1.4911	0.6706
Urban Area Distance	1.4112	0.7082
Unemployment Rate	1.1471	0.8718
Terrain Slope	1.0271	0.9736
Market Revenue	1.0976	0.9111

Table 2: Variance Inflation Factor (VIF) and Tolerance

Panel of Figure 4 — Figure 4: Feature importance Scores by five ML Classifiers

Classifier	Accuracy	Recall	Precision	F1 Score	HSC	SC	SSC	NSC	AUC
RF	0.8648	0.8648	0.8734	0.8663	0.9119	0.8017	0.8530	0.9658	0.9656
XGBoost	0.8495	0.8495	0.8588	0.8511	0.9080	0.9596	0.8424	0.7620	0.9569
SVC	0.8186	0.8186	0.8286	0.8199	0.9157	0.7035	0.8172	0.9161	0.9159
LR	0.5870	0.5870	0.6919	0.5862	0.8812	0.5741	0.4460	0.9068	0.8597
KNN	0.7257	0.7257	0.7600	0.7313	0.8008	0.6618	0.6962	0.8727	0.8142

Table 3: Performance comparison. HSC: Highly Suitable Class, SC: Suitable Class, SSC: Somewhat Suitable Class, NSC: Not Suitable Class.

Panel of Figure 5 — Figure 5: Confusion Matrix of five ML Classifiers

Panel of Figure 6 — Figure 6: AUCROC Curve of five ML Classifiers

Model	Accuracy	Recall	Precision	F1 Score	HSC	SC	SSC	NSC	AUC
RF (9 Features)	0.8404	0.8404	0.8529	0.8430	0.8621	0.7912	0.8278	0.9441	0.9629
RF (7 Features)	0.8648	0.8648	0.8734	0.8663	0.9119	0.8017	0.8530	0.9658	0.9656

Table 4: Comparison of Model Performance with 9 and 7 Features using the RF model

Panel of Figure 7 — Figure 7: Impact of individual features on the output of five machine learning classifiers (RF, XGB, SVC, LR, and KNN) using SHAP analysis. Each plot visualizes the contribution of features to the model predictions. The color scale represents the feature value (red for high and blue for low), while the SHAP values on the x-axis indicate the magnitude and direction of each feature’s influence on the model output.

Panel of Figure 8 — Figure 8: Mean SHAP values of the top features across five machine learning classifiers (RF, XGB, SVC, LR, and KNN). The horizontal bars represent the average absolute SHAP values, quantifying each feature’s overall contribution to model predictions.

Adjusted Weights	Road Dist.	Rail Line Dist.	Urban Dist.	Unemp. Rate	Market Rev.	SDR	Precipit.
Baseline	0.143	0.143	0.143	0.143	0.143	0.143	0.143
RF	0.195	0.182	0.156	0.117	0.091	0.195	0.065
XGB	0.182	0.182	0.156	0.117	0.091	0.208	0.065
SVC	0.189	0.149	0.203	0.122	0.054	0.216	0.068
LR	0.213	0.190	0.204	0.101	0.077	0.176	0.040
KNN	0.192	0.128	0.192	0.128	0.026	0.218	0.115

Table 5: Comparison of Adjusted Weights

Figure 9: Comparison of Adjusted Weights by Different ML Classifiers

Panel of Figure 10 — Figure 10: Suitability Map of Mississippi using five ML models

Figure 11: Area distribution of 4 Suitability Classes using RF, KNN, LR, XGB, and SVC model

Figure 12: Existing Sawmill Distribution over the different Suitability Class

Multi-Criteria Decision Analysis

Multi-Criteria Decision Analysis (MCDA) ranks or selects alternatives by balancing competing criteria, preferences, uncertainty, and constraints.

See all 35 papers

A Remote Approach to Cashew Orchard Detection: Leveraging Active Learning with Satellite Imagery in Guinea-Bissau
Aug 12, 2026Miguel Pereira, Sofia C. Pereira, Maria J. P. Vasconcelos +3Agricultural Remote Sensing
When Does Machine Learning Beat Value Sorting? A Three-Dataset Diagnostic of Exposure-Weighted Shipment Prioritization
Jul 20, 2026Jize LiLearning to Rank Supply Chain Management
Boston University, United States
Technical Report on the Turba Fertilizer Machine Learning Stack in Morocco
Oct 5, 2026Abdelghani Belgaid, Zakaria Mahmoud, Fahd Chibani +3Surrogate Modeling ML Reproducibility
Turba Open Lab

Aug 12, 2026J/K move · Enter open · S save

A Remote Approach to Cashew Orchard Detection: Leveraging Active Learning with Satellite Imagery in Guinea-Bissau

Miguel Pereira, Sofia C. Pereira, Maria J. P. Vasconcelos, Patrícia Guedes+2

Agricultural Remote Sensing

Open paper AskPDF arXiv

Cashew production is a widespread economic activity in Guinea-Bissau, as well as other countries in West Africa. However, unregulated cashew production can be directly associated with increasing regionwide deforestation rates, biodiversity losses, and a fragile economic structure. There is no nationwide database for listing or georeferencing cashew orchards, so there is a clear need to remotely map their locations. In recent years, multiple methods for detecting orchards have been developed, though they have only been applied on a regional level. This work expands regional analyses to a nationwide scale. It develops a scalable and cost-effective remote approach, based on Sentinel-2 satellite imagery, using Machine Learning techniques to detect cashew orchards automatically. Margin-based Active Learning techniques were employed to develop an optimal training set in terms of the number of points and their informativeness, leading to a cashew map with 94.0% balanced accuracy obtained entirely off-site. We created two datasets and a 2021 cashew map with 10m spatial resolution that are openly accessible through GitHub. The results demonstrate the possibility of a broader cashew orchard mapping, creating a new stepping stone for this environmental application.

Abstract

Multi-Criteria Decision Analysis

Explore similar work

A Remote Approach to Cashew Orchard Detection: Leveraging Active Learning with Satellite Imagery in Guinea-Bissau

When Does Machine Learning Beat Value Sorting? A Three-Dataset Diagnostic of Exposure-Weighted Shipment Prioritization

Technical Report on the Turba Fertilizer Machine Learning Stack in Morocco