Scholarlink

<p>Dynamic analytical frameworks are essential for socioeconomic vulnerability assessment in post-transition economies, particularly for capturing temporal evolution and identifying actionable policy thresholds. This study integrates time-varying panel regression with explainable machine learning to analyze multidomain vulnerability patterns across Visegrád countries using harmonized Eurostat data from 2015 to 2024 (n = 40 country observations). The methodology employs composite vulnerability indices from six domains (exposure, sensitivity, capacity, economic, housing, infrastructure), gradient boosting models with partial dependence analysis, and counterfactual policy scenarios. Time-varying panel regression achieves R<sup>2</sup> = 0.984 with significant temporal coefficient evolution. Sensitivity domain effects increase from 0.283 to 0.528 (<i>p</i> &lt; 0.001). Explainable machine learning reveals&#xa0;overcrowding rate as the dominant vulnerability predictor (81.8% feature importance), dramatically overshadowing GDP per capita (1.3%), with unmet medical needs exhibiting secondary influence (9.2%). Partial dependence analysis identifies critical policy thresholds at standardized values of − 0.03 for overcrowding and − 0.71 for unmet medical needs, revealing threshold-dominated nonlinear relationships rather than linear proportional effects. Country-specific trajectories show Hungary achieving largest historical improvement (− 0.924 units), Czechia maintaining strongest final position (− 1.292), and Poland exhibiting highest remaining vulnerability (0.234 in 2024). Counterfactual intervention scenarios demonstrate mean regional vulnerability reduction of − 0.426 units through combined housing, healthcare, and poverty interventions, with overcrowding improvements alone yielding&#xa0;0.334 units reduction. Country-specific intervention responsiveness ranges from minimal (− 0.004 for Czechia) to substantial&#xa0;(− 0.766 for Poland and − 0.751 for Slovakia), revealing that elevated baseline vulnerabilities predict stronger intervention effectiveness. The integrated framework advances evidence-based regional development strategies prioritizing housing quality and healthcare access.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable machine learning for multidomain socioeconomic vulnerability assessment in the Visegrád region

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月3日

查看全文
  • Mohammad Fazle Rabbi

摘要

Dynamic analytical frameworks are essential for socioeconomic vulnerability assessment in post-transition economies, particularly for capturing temporal evolution and identifying actionable policy thresholds. This study integrates time-varying panel regression with explainable machine learning to analyze multidomain vulnerability patterns across Visegrád countries using harmonized Eurostat data from 2015 to 2024 (n = 40 country observations). The methodology employs composite vulnerability indices from six domains (exposure, sensitivity, capacity, economic, housing, infrastructure), gradient boosting models with partial dependence analysis, and counterfactual policy scenarios. Time-varying panel regression achieves R2 = 0.984 with significant temporal coefficient evolution. Sensitivity domain effects increase from 0.283 to 0.528 (p < 0.001). Explainable machine learning reveals overcrowding rate as the dominant vulnerability predictor (81.8% feature importance), dramatically overshadowing GDP per capita (1.3%), with unmet medical needs exhibiting secondary influence (9.2%). Partial dependence analysis identifies critical policy thresholds at standardized values of − 0.03 for overcrowding and − 0.71 for unmet medical needs, revealing threshold-dominated nonlinear relationships rather than linear proportional effects. Country-specific trajectories show Hungary achieving largest historical improvement (− 0.924 units), Czechia maintaining strongest final position (− 1.292), and Poland exhibiting highest remaining vulnerability (0.234 in 2024). Counterfactual intervention scenarios demonstrate mean regional vulnerability reduction of − 0.426 units through combined housing, healthcare, and poverty interventions, with overcrowding improvements alone yielding 0.334 units reduction. Country-specific intervention responsiveness ranges from minimal (− 0.004 for Czechia) to substantial (− 0.766 for Poland and − 0.751 for Slovakia), revealing that elevated baseline vulnerabilities predict stronger intervention effectiveness. The integrated framework advances evidence-based regional development strategies prioritizing housing quality and healthcare access.