Application of Stacked Machine Learning Models in Population Development Equations for Population Forecasting

The current changes in China’s population structure and dynamics have led to profound challenges in population planning, forecasting, decision-making, and early warning. To address the issues of predicting age- and gender-specific population retention, migration, and birth rates, a combination model of Multilayer Perceptron (MLP) and Random Forest (RF) is constructed using stacking techniques, with a discrete population development equation as the base model. The MLP-RF model is employed to perform regression training on population data, resulting in a novel ensemble approach to population forecasting. The study uses the data from the sixth and seventh national censuses of Hebei Province, reconstructing population data for 2010-2020. After data training and error evaluation, it is demonstrated that the ensemble forecasting model has excellent predictive capabilities for population retention, migration, and birth-related issues.