GRU-based forecasting of conflict-related socio-economic vulnerabilities under illicit practices

  • 11 Views
  • 2 Downloads

Creative Commons License DMCA.com Protection Status
This work is licensed under a Creative Commons Attribution 4.0 International License

Type of the article: Research Article

Abstract
The study develops a framework for forecasting conflict-related socio-economic vulnerabilities in a cross-country sample using indicators of armed conflict risk, forced displacement, and illicit practices. The analysis uses panel data covering 135 countries over 2012–2024. The Conflict Risk Index, Refugee Load Index, and Cyber Vulnerability Index were constructed using standardized indicators and Principal Component Analysis. The Cyber Vulnerability Index is a proxy for digital exposure to cyber-related disruptions. First-difference panel regressions identified associations between conflict- and migration-related risks and socio-economic indicators. Separately, GRU neural networks were applied to forecast these indicators, while comparative quartile-based ablation and scenario-based perturbation analyses assessed the contribution and sensitivity of corruption, AML-related risk, and cyber vulnerability across country groups. Regression analysis showed that the Conflict Risk Index was statistically associated with deterioration in GDP, GDP per capita, GDP per capita growth, political stability, life expectancy, labor force dynamics, and migration balance. The Refugee Load Index was associated with positive net migration alongside weaker economic growth and lower political stability in host countries. Forecasting results were heterogeneous: adding corruption, AML-related risk, and cyber vulnerability indicators improved accuracy for several targets, but deteriorated performance for some variables and showed evidence of overfitting for GDP per capita in the conflict-risk specification. In high-conflict-risk countries, the largest forecast improvements were for life expectancy (61.02%), GDP per capita (43.96%), and net migration (10.75%); in high-refugee-load countries, they were observed for life expectancy (60.30%), political stability (38.41%), and net migration (16.20%). The framework can support early warning, resilience assessment, and crisis response.

Acknowledgment
We acknowledge with gratitude the financial support provided by the Ministry of Education and Science of Ukraine for the research project “Cybersecurity and digital transformations of the country’s wartime economy: the fight against cybercrime, corruption and the shadow sector,” state registration number 0124U000544.

view full abstract hide full abstract
    • Figure 1. Spatial differentiation of countries worldwide by the Conflict Risk Index in: a) 2012; b) 2014; c) 2015; d) 2020; e) 2022; f) 2023
    • Figure 2. Spatial differentiation of countries worldwide by the Refugee Load Index in: a) 2012; b) 2014; c) 2015; d) 2020; e) 2022; f) 2023
    • Figure 3. Proxy feature importance in baseline GRU models with the Conflict Risk Index
    • Figure 4. Proxy feature importance in extended GRU models with the Conflict Risk Index
    • Figure 5. Proxy feature importance in baseline GRU models with the Refugee Load Index
    • Figure 6. Proxy feature importance in extended GRU models with the Refugee Load Index
    • Figure 7. Comparative quartile-based ablation analysis for countries with high and low (a) conflict risk and (b) refugee load
    • Figure 8. Scenario-based perturbation analysis under 1%, 5%, and 10% increases in AML, corruption risks, and cyber vulnerability for countries with high and low conflict risk
    • Figure 9. Scenario-based perturbation analysis under 1%, 5%, and 10% increases in AML, corruption risks, and cyber vulnerability for countries with high and low refugee load
    • Table 1. List of indicators
    • Table 2. Panel regression results for socio-economic indicators
    • Table 3. Machine learning results of baseline and extended GRU models
    • Conceptualization
      Hanna Yarovenko, Olena Pakhnenko
    • Data curation
      Hanna Yarovenko
    • Funding acquisition
      Hanna Yarovenko, Olena Pakhnenko, Liudmyla Riabushka
    • Investigation
      Hanna Yarovenko, Olena Pakhnenko, Liudmyla Riabushka
    • Methodology
      Hanna Yarovenko, Viktoriia Khmurova
    • Project administration
      Hanna Yarovenko, Olena Pakhnenko
    • Supervision
      Hanna Yarovenko
    • Visualization
      Hanna Yarovenko, Liudmyla Riabushka
    • Writing – original draft
      Hanna Yarovenko, Olena Pakhnenko, Viktoriia Khmurova
    • Writing – review & editing
      Hanna Yarovenko, Liudmyla Riabushka, Iryna Tarasenko
    • Resources
      Olena Pakhnenko, Liudmyla Riabushka, Iryna Tarasenko
    • Validation
      Liudmyla Riabushka, Iryna Tarasenko, Viktoriia Khmurova
    • Formal Analysis
      Iryna Tarasenko, Viktoriia Khmurova