A machine learning framework for forecasting multidimensional sustainability and informing integrated policy thresholds in the EU

Abstract Achieving sustainable development in the European Union requires integrated frameworks capable of navigating tradeoffs among economic growth, social equity, and environmental resilience. This study develops an advanced data-driven forecasting and policy optimization framework, integrating multidimensional sustainability indicators for seven EU countries. Using EUROSTAT data from 2014 to 2023, the framework generates forecasts up to 2030 to support strategic policy insights. The methodology combines Random Forest regression for feature selection, Long Short-Term Memory networks for temporal forecasting, SHapley Additive exPlanations for interpretability, and the Technique for Order of Preference by Similarity to Ideal Solution for multi-criteria decision analysis. Three principal drivers emerge: Global Value Chain Backward Participation, Social Protection Expenditure, and Municipal Recycling Rate. Critical policy thresholds are identified: Foreign Direct Investment below €327 million balances economic gains and environmental costs; social protection spending above 25.6% of Gross Domestic Product reduces workplace fatalities when the gender pay gap remains under 21.3%; and a 10% increase in recycling rates correlates with a 2.3 tonnes per capita reduction in carbon dioxide emissions. LSTM projections forecast GDP stabilization at 2% to 3% by 2030 but highlight resource productivity declines in Southern Europe without targeted intervention. Country rankings position Germany, France, and Italy as sustainability leaders according to the multi criteria analysis. The framework translates these insights into a dynamic policy matrix encompassing tiered FDI incentives, AI-driven welfare systems, and blockchain-based supply chain tracking. This approach bridges predictive analytics and evidence-based policy design for EU sustainability governance.