Logo image
Unveiling the ESG weighting system using multi-level global analysis: A machine learning approach with SHAP explainability
Journal article   Peer reviewed

Unveiling the ESG weighting system using multi-level global analysis: A machine learning approach with SHAP explainability

Leili Soltanisehat, Omkar Annaray Badadale and Nefeli Bompoti
Journal of cleaner production, Vol.571, 148811
07/08/2026

Abstract

ESG Industrial context Machine learning Regional context SHAP value
Environmental, Social, and Governance (ESG) metrics are increasingly critical for evaluating corporate sustainability and investment risk. As of today, several ESG frameworks consider different metrics, often assigning different weights to ESG pillars. However, these weights are often derived through opaque methodologies and could vary at the industry and company levels, limiting transparency and cross-company comparability. This research seeks to understand the primary drivers of ESG's weight emphasis across firms while maintaining regional and industry contextualization. In addition, interpretable models are developed using machine learning that predict ESG pillar weights based on structured economic and market variables, while considering the industry- and region-level heterogeneity among firms. The analysis dataset consisted of 1327 observations spanning 13 industrial groups and 27 countries in three regions of Asia, Europe, and North America. The study provides insights into how financial performance, industry concentration, and, to a lesser extent, regional context influences the weighting of the ESG pillars. Our findings demonstrate the need for tailored ESG frameworks that capture the firm-specific and industry-specific components while accounting for regional-level considerations.

Metrics

1 Record Views

Details

Logo image