Forecasting and Analyzing the Drivers of Rural Residents' Income Using Stacking Ensemble Learning: A Case Study of Zhuji City

Abstract

Forecasting rural income is vital for China's national strategies, yet traditional models falter on short, multi-feature time-series data, creating an accuracy-interpretability dilemma. This paper introduces a Stacking ensemble learning framework using 2012-2021 economic data from Zhuji City to address this. The framework integrates Ridge, SVR, and Random Forest base learners, with a Linear Regression meta-learner to optimally weigh their predictions. Empirical results demonstrate two key findings: 1) The Stacking model achieves significantly superior predictive accuracy over its base learners, proving its robustness on small-sample data. 2) An analysis of the meta-learner weights reveals that rural income growth follows a composite pattern, dominated by a linear trend and fine-tuned by nonlinear corrections. This research offers a robust forecasting paradigm for regional economies and a novel, interpretable method for analyzing the driving mechanisms of complex economic phenomena.