DUAL-MODE EXPLAINABLE AI SYSTEM FOR ESG GREENWASHING DETECTION: COMPANY-LEVEL AND CUSTOM TEXT ANALYSIS WITH SHAP TRANSPARENCY

Abstract

This study introduces a Dual-Mode Explainable AI (XAI) System for the identification of ESG (Environmental, Social, and Governance) greenwashing in corporate sustainability reports. The system employs Natural Language Processing (NLP) with TF-IDF features and Logistic Regression to analyse the disclosures of companies and a text provided by a user. By using SHAP (SHapley Additive exPlanations) and its LinearExplainer, the system is very clear about the pieces of information that influence the prediction. The dual architecture comprises analyses at the Company-Level and of a Custom Text, thus ensuring interpretability at both the global and local levels. The model is very close to being an ideal one in terms of class weights and thus the system has a strong performance capability while at the same time it is able to convert complicated predictions into pieces of evidence that are easily understood by human beings and thus it can be said that the system is quite helpful in bringing about regulatory transparency and investor trust.