Akademik Veri Yönetim Sistemi
FRONTIERS IN SUSTAINABILITY, cilt.7, ss.1-23, 2026 (ESCI, Scopus)
Introduction:
Sustainable supply chain performance cannot be assessed only by looking at one group of indicators. It needs a framework that can bring environmental, social, and organizational factors together with operational outcomes. Traditional multi-criteria decision-making methods usually depend on fixed weights and mostly linear logic. This makes them weak when sustainability relations are nonlinear, or when variables affect each other in interaction form. For this reason, this study proposes an explainable machine learning framework to assess sustainable supply chain performance.
Methods:
The framework applies dual-model architecture. Random Forest regression is used for prediction of continuous Sustainability_Score, and XGBoost classification is used for grouping firms into sustainability categories. The dataset contains environmental indicators, organizational and human-capital variables, also structural indicators related to supply chain. Sustainability performance is measured by a composite ESG-based index. After that, this index is transformed into ordinal classes. SHAP analysis is used to explain effects of features and also interactions between them. The stability of these explanations is checked across cross-validation folds. The framework is also compared with baseline machine learning models and with TOPSIS, as one traditional MCDM method.
Results:
A holdout test set was used to evaluate the proposed framework. The Random Forest regression model reached R2 = 0.9094 on this test set. The XGBoost classification model also reached AUC = 0.9565. The SHAP values show that sustainability performance of firms is mainly driven by emissions intensity, organizational trust, and employee participation. The SHAP stability test also shows that feature importance remains stable across validation folds.
Discussion:
The findings show that sustainable supply chain performance is not made only by environmental factors. Social and organizational conditions also play important role in this performance. Based on interaction analysis, social factors become more visible and more influential when emissions are at lower level. Therefore, the proposed framework can be considered as an interpretable and data-driven alternative for fixed-weight MCDM approaches. It can also support ESG-oriented strategies, supplier evaluation, and procurement decisions which have sustainability focus.