DATA-DRIVEN GREEN SUPPLIER DEVELOPMENT: CARBON IMPACT ANALYSIS WITH RANDOM FOREST AND SHAP

The relevance of sustainability in supply chain management has increased interest in methods capable of supporting the evaluation and development of suppliers based on environmental criteria. Although there are studies on the evaluation and development of green suppliers, evidence on the relationship between green criteria and carbon impact is still limited. Thus, this work aimed to identify which green criteria most influence the classification of the carbon impact of suppliers, combining predictive modeling, SHAP-based interpretability and classical statistical analyses, in addition to proposing directions for the development of suppliers based on the literature. The survey was conducted with data from 50 suppliers evaluated on 15 green criteria. Random Forest, Random Search, Grid Search, repeated holdout validation, SHAP, correlations, and logistic regression were used. It is concluded that the approach adopted can make the evaluation and development of suppliers more targeted, despite the sampling limitations.