Population Density Prediction at Township Scale Supported by Machine Learning Method: A Case Study in Inner Mongolia

With the acceleration in population migration and urbanization, accurate population density prediction has become increasingly important for regional planning and resource management. This study focuses on predicting population density at the township level in Inner Mongolia. By integrating multi-source data, such as nighttime light indices and road network density, various machine learning models—including random forest, XGBoost, and LightGBM—were employed to significantly improve prediction accuracy. Interpretable machine learning techniques were utilized to quantitatively analyze the contribution of various variables to population distribution. The results indicate that nighttime light indices and road network density are key influencing factors, revealing their complex nonlinear relationships with population density. This study provides new methodological support for predicting population density in Inner Mongolia and similar regions, demonstrating the potential of machine learning in regional population research. While machine learning models effectively capture correlations between variables, they do not reveal causal relationships. Future research should introduce more detailed data and causal inference models to deepen our understanding of population distribution and its influencing factors.