Intelligent Identification of Regional Economic Growth Driving Factors and Nonlinear Explanatory Model Using LightGBM

Abstract

The identification of driving factors of regional economic growth is often limited by the assumptions of traditional linear models, which makes it difficult to capture the complex nonlinear relationship and characteristic interaction effects of multi-factor spatial spillover and regional heterogeneous evolution. This study aims to construct a novel method based on Light Gradient Boosting Machine (LightGBM) to identify the driving factors of regional economic growth. The intelligent analysis framework accurately identifies key driving factors and quantifies their nonlinear impact mechanisms from a multi-scale perspective of regional coordinated development. First, it integrates multi-source heterogeneous data to construct an 81-dimensional regional economic feature set (covering factor endowment, spatial correlation and institutional environment dimensions), and uses adaptive feature cross-talk technology to analyze the factor synergy network. It uses the histogram optimization algorithm and Leaf-wise growth strategy of LightGBM to accelerate training, optimizes hyperparameters through Bayesian optimization, and introduces SHAP values for cross-regional feature attribution, and combines local dependency graphs to reveal the gradient effect under the core-edge structure. In the verification of 283 prefecture-level cities, the model R2 reached 0.912. Key findings include: the digital economy development index presents an inflection point of increasing marginal benefits in the metropolitan area, the spatial synergy effect contribution rate of human capital interaction terms and FDI reaches 18.3%, and traditional infrastructure investment presents an inverted U-shaped contribution due to differences in regional development stages. The LightGBM-SHAP framework breaks the "geographic paradox"of the regional growth dynamics system and provides a data-driven decision-making basis for the differentiated regional policy system.