Bayesian Optimization in Hybrid Data Envelopment Analysis and Stacking Approach for Optimizing Supplier Selection

This study addresses the prescriptive limitations of traditional Data Envelopment Analysis (DEA) by proposing a novel hybrid framework, the Hybrid DEA-Stacking-Bayesian Model, to enhance supply chain optimization. An input-oriented DEA model under Variable Returns to Scale (VRS) is employed to minimize inputs while maintaining output levels. The efficiency scores obtained from the DEA model serve as outputs for machine learning models trained to predict the efficiency of decision-making units based on their input-output profiles. The hybrid approach integrates DEA with Stacking algorithms and incorporates Bayesian optimization for hyperparameter tuning. Results demonstrate that Bayesian optimization significantly enhances performance across all meta-learners, with XGBoost achieving the highest accuracy, improvement in R2, and yielding the lowest error metrics. Comparative analyses highlight XGBoost as the best-performing model, followed by kNN, RF, and DT, confirming the effectiveness of the proposed framework. The integration of DEA and Stacking improves supplier selection by evaluating efficiency and enhancing prediction accuracy through model combination. This work provides a robust decision-support tool for supply chain management, laying the groundwork for future research on supplier selection.