A hypothesis-driven responsible AI framework for interpretable ESG forecasting with RuleFit

Area of Science:

• * Business Analytics
• * Artificial Intelligence
• * Corporate Social Responsibility

Background:

• * Responsible AI (RAI) is integral to corporate social responsibility and business governance.
• * Predicting Environmental, Social, and Governance (ESG) scores is crucial for sustainable finance.
• * Existing methods may lack interpretability and ethical alignment in ESG forecasting.

Purpose of the Study:

• * To develop a framework integrating RAI principles with hypothesis-driven rule extraction for ESG score prediction.
• * To enhance the interpretability and ethical alignment of ESG outcome forecasts.
• * To facilitate reliable decision-making in sustainable finance and corporate governance.

Main Methods:

• * Employed the ensemble-based RuleFit algorithm for hypothesis-driven rule extraction.
• * Associated firm-level attributes (size, leverage, digitization, ownership) with decision rules.
• * Validated rules using nonparametric tests and assessed subgroup fairness for equity.

Main Results:

• * Generated interpretable decision rules linking firm attributes to ESG outcomes.
• * Achieved statistically sound and ethically aligned ESG score predictions.
• * Demonstrated the robustness of the methodology under non-normal distributions.

Conclusions:

• * Rule-based learning offers a powerful approach for interpretable and ethical ESG forecasting.
• * The proposed framework supports data-driven, responsible decision-making in corporate governance.
• * This research bridges AI, ESG, and sustainable finance through a novel methodology.