Abstract: The increasing relevance of Environmental, Social, and Governance (ESG) factors in portfolio optimization requires advanced methodologies capable of handling data heterogeneity and non-linear interactions. This study introduces the Dynamic Ensemble Learning Strategy (DELS), a machine learning-based framework that enhances firm selection for sustainable portfolio construction. DELS integrates a modified Random Forest algorithm ensuring an adaptive decision making process based on a two-stage consensus mechanism, optimizing risk-adjusted returns while aligning with sustainability objectives. The methodology is applied to Euro STOXX 600 stocks over the period from 2021 to 2024, demonstrating improved robustness against market fluctuations and enhancing ESG adherence. By combining different splitting measures, we also propose a volatility based recalibration strategy able to outperform iShares STOXX Europe 600 UCITS ETF and a passive buy-and-hold investment based on DELS, which demonstrates the practical implications of the model into Socially Responsible Investing.

https://link.springer.com/article/10.1007/s10479-026-07459-5