Synergistic Meta-Learning for Sustainable Finance: A Hybrid Deep-Tree Fusion Architecture for ESG Score Prediction

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:ESGスコア予測における深層学習とツリーアンサンブルの融合手法の有効性を示す実証的知見を提供。

🏢実務担当者:ESG評価の精度向上により、投資判断やリスク管理の改善に活用可能。

🏛政策担当者:ESGデータの信頼性向上に寄与し、開示規制の実効性を高める可能性。

Environmental, Social, and Governance (ESG) performance is now a core part of corporate stability, but traditional financial measures are no longer sufficient to assess value amidst rising environmental and social pressures. ESG data remains complex and is often criticized for being opaque or prone to greenwashing. This methodological challenge is exacerbated by the fact that standard deep learning models do not always outperform tree ensembles on structured data, leaving a gap in reliable forecasting. This study develops the Deep-Tree Fusion framework to address these issues by synergistically integrating tree-based logic with differentiable deep learning. The study uses a long-term dataset of 11,000 firm-years spanning 2015-2025. The architecture follows a hierarchical design that first extracts non-linear structures from tabular data and then refines them through a specialized fusion network. Time-based integrity is maintained by sorting data chronologically to avoid the biases of random sampling. Empirical findings demonstrate that DTF achieves a superior predictive fit ( R 2 = 0.97866) and significantly outperforms 10 conventional baseline models. Practically, this work provides a reliable diagnostic tool for investors and policymakers to measure sustainability-adjusted value and manage long-term risks. The framework is robust across various industries and maintains stable performance despite market volatility.

• openalex https://doi.org/10.1016/j.joitmc.2026.100851first seen 2026-08-14 05:07:28