A dual-stage machine learning framework versus traditional OLS for forecasting regional economic resilience through industry investment and tourism income

Sectoral fixed assets investment influences regional economic growth through tourism resilience, which has undergone significant changes in the post-pandemic world. This study develops a dual-stage machine learning framework to forecast the transmission mechanism from fixed asset investment to regional GDP, with domestic tourism income as an intermediary variable, using panel data from 31 Chinese provinces between 2012 and 2023, integrating log-linear regression, Random Forest (RF), and LightGBM (LGBM) models to compare predictive performance and capture nonlinear effects, considering structural breaks associated with the COVID-19 pandemic to examine system shocks and transitions in investment efficiency. It is revealed that real estate, transportation, wholesale and retail trade investments exhibit the highest contributions to tourism and GDP, with a shift in system responsiveness after 2020 and becoming the forecast basis for forward-looking decisions.