A Modular Decision Support System for Economic Resilience: Integrating Machine Learning and Policy Alignment

The 2007–2009 financial crisis exposed the limits of Greek regional policy: GDP fell 24.4% between 2008 and 2016 across the 13 regions. We present a Decision Support System (DSS) for economic resilience that brings together a composite index, machine-learning forecasts, policy analytics, sectoral signals and EU AI Act safeguards in four Python modules. Module ΠΕ1 builds a Composite Resilience Index (CI) for 2008–2024 from the Martin Index and shift-share decomposition, with dual weighting via Shannon entropy and SHAP importance; admissibility is governed by a bootstrap stability gate (B = 500 resamples, 95% SHAP-weight CI half-width < 5 pp, bootstrap mean within 10 pp of the Shannon baseline). Module ΠΕ2 trains a Vulnerability Index (VI) on quarterly data (a panel of 988 observations, 13 regions × 76 quarters, 2006Q1–2024Q4) to forecast 2025–2026. Module ΠΕ3 derives a Policy-Index Alignment Score (PIAS) from ESPA programming documents through retrieval-augmented generation, together with a Funding-per-Need Index. Module ΠΕ4 synthesises CI, (1−VI) and PIAS into a DSS score using entropy-derived weights, in line with the transparency and record-keeping provisions of the EU AI Act. In the pilot run, South Aegean leads on CI (0.7309) while Western Greece, Epirus and Western Macedonia score lowest. The 2025 VI projections place four regions in RED, six in YELLOW and three in GREEN. In 2026 the DSS weights prioritise vulnerability [w(1−VI) = 0.7414], which favours South Aegean (0.9392) and the Ionian Islands (0.8492) and leaves Epirus (0.1098) exposed under the μ±0.5σ classification. External validation against the EU Regional Competitiveness Index (RCI) indicates that the DSS measures something other than competitiveness. In the counterfactual tests, reducing vulnerability raises DSS scores far more than improving alignment. For policy makers the DSS provides rankings, early warnings, alignment gaps and sectoral signals.