Economic Health Risk Detection Mechanisms by Super-Resolution Neural Network Algorithms

This study addresses the limitations of traditional macroeconomic risk monitoring in capturing subtle structural changes through image-based analysis. Using public databases such as the World Bank's World Development Indicators (WDI) and OECD Health Statistics, key “economic vital signs” indicators are selected. These include the Gross Domestic Product (GDP) growth rate, the unemployment rate, the ratio of household credit to GDP, and healthcare expenditure as a percentage of GDP. Multidimensional time series are constructed for G20 countries from 2000 to 2022. Time-series-to-image encoding methods, including Gramian Angular Fields (GAF) and Recurrence Plots (RP), convert macroeconomic data into economic state maps. Samples are generated using a five-year sliding window, and only windows for which the next-year risk label is observable are retained. Within the 2000–2022 study period, the valid supervised windows range from 2000–2004 to 2017–2021, resulting in a total of 342 labeled economic state map samples. The 2018–2022 window corresponds to the 2023 risk state. Because 2023 falls outside the temporal scope of this study, this window is excluded from model training, supervised testing, and statistical evaluation. Under the annual data frequency, one-year-ahead prediction is equivalent to an economic health risk early-warning task conducted four quarters in advance. Based on the economic state map, a Super-Resolution Economic State Network (SR-EconomicNet) is developed. The model takes the GAF channel and the RP channel as inputs to a shallow convolutional layer. High-frequency textures are then reconstructed through Residual Channel Attention Blocks, followed by sub-pixel upsampling to generate a high-resolution economic image. Compared with the conventional Super-Resolution Residual Network (SR-ResNet), SR-EconomicNet aims to achieve pixel-level reconstruction while incorporating pressure concentration regions, recurrence breakpoints, and trend distortions into the reconstruction objective. This design improves the visibility of weak signals that emerge prior to crisis events. Key historical periods, such as the Global Financial Crisis, the European Debt Crisis, and the Coronavirus Disease 2019 pandemic, serve as evaluation intervals. Using G20 panel data, the proposed method is compared with alternative models. The results show that SR-EconomicNet achieves higher mean performance than the comparison models in pre-crisis early-warning tasks. Its prediction accuracy for economic recessions reaches 0.873, exceeding Vector Autoregression by 0.095 points. The model also captures high-frequency risk signals more effectively, revealing clearer paths of risk accumulation before shocks. Introducing super-resolution image reconstruction into economic health risk detection enhances early identification of macroeconomic risks. This study provides a technical foundation for higher-resolution, more sensitive risk monitoring systems and supports policymakers in implementing cross-sectoral interventions during the early stages of risk accumulation.