게이트순환유닛(GRU)–릿지 회귀 하이브리드 모델 기반 제조업 창업 예측 및 XAI 기반 해석

This study employs a hybrid model combining Gated Recurrent Unit (GRU) and Ridge regression to forecast trends in manufacturing firm births in South Korea and to interpret their determining factors. The GRU model, which captures temporal dependencies, was used to forecast the number of manufacturing firm births, followed by residual correction using Ridge regression with L2 regularization, enhancing both forecasting accuracy and stability. To ensure model interpretability, SHapley Additive exPlanations (SHAP) analysis was conducted. Monthly data on key macroeconomic and industrial indicators, including interest rates, exchange rates, trade balance, Business Survey Index (BSI), and the Small and Medium Manufacturing Production Index, were utilized. The last 12 months of the time series were reserved for validation, and the preceding period was used for model training. The results indicate that the proposed hybrid model outperforms the GRU-only model in forecasting performance. SHAP analysis revealed that exchange rates, the Small and Medium Manufacturing Production Index, and BSI contribute most significantly to manufacturing firm births forecasting. Furthermore, increases in exchange rates and interest rates consistently exert a negative effect on the forecasted number of manufacturing firm births. This study provides an integrated perspective on the nonlinear effects of complex macroeconomic factors on manufacturing firm births, offering valuable insights for entrepreneurship policy design and industrial support strategies.