A Risk-Integrated Multi-Objective Optimization Framework for Dynamic Contractor Allocation in Zimbabwe's Commercial Timber Value Chain

Abstract

Zimbabwe's commercial sector in forestry is currently using small and medium scale contractors whereas as much as 70% of harvesting and milling is outsourced, resulting in chronic supply chain inefficiencies that previous Enterprise Resource Planning (ERP) systems cannot address. In this paper, we present a Dynamic Resource Allocation Framework (DRAF) for bridging the decision-intelligence gap by feeding the XGBoost-derived contractor risk probabilities directly into a Non-dominated Sorting Genetic Algorithm II (NSGA-II) multi-objective optimizer to allow estate managers to obtain the Pareto-optimal contractor-block-mill assignment that simultaneously minimizes cost, maximizes expected timber recovery, minimizes expected delay, maximizes operational reliability, and minimizes transport distance. The proposed framework was developed and evaluated on an 828,789-record Virtual ERP dataset built on the USDA Forest Service Timber Harvests Feature Layer and calibrated to Manicaland's forestry areas; no proprietary Zimbabwean ERP records were available, so the results are interpreted as synthetic benchmark evidence rather than production validation. XGBoost performed better than (ROC-AUC 0.600) Logistic Regression baseline on the holdout ROC-AUC of 0.965, recall of 0.999, and F1 of 0.867. The NSGA-II optimizer gave 64 fully feasible Pareto solutions for the balanced and high-recovery cases and identified a constraint-feasibility threshold for the strict, low-risk scenario, articulating a suitable operational trade-off. An interactive Dash decision-support prototype was functionally demonstrated across all evaluated pipeline components and used to present the framework's outputs to non-technical estate managers. A structured post-demonstration stakeholder evaluation with 30 practitioners produced a mean formative satisfaction score of 4.19/5.0. Field usability testing, real-data validation, and a formal ablation of the ML-risk integration remain necessary future validation steps. As an extension to the research, this research is also known as the first integrated, domain-specific decision-intelligence framework for forestry contractor allocation in Sub-Saharan Africa in this context and provides a replicable model for best practice, methodological framework for risk-integrated multi-objective optimization in resource and other intensive industries.