Analysis of the Factors Related to the Total Fertility Rate and the Rate of Increase in Municipalities in Japan Using Machine Learning

Japan faces a severe demographic challenge marked by a declining birth rate and an aging population. This study investigates the regional factors influencing total fertility rate (TFR) and its recent changes across Japanese municipalities by applying machine learning techniques, particularly the LightGBM algorithm. Using data from national censuses and economic surveys, we constructed cross-sectional and time-series models to estimate TFR in 2022 and its change from 2017. The machine learning models significantly outperformed traditional linear regression methods, achieving higher explanatory power (R 2 = 0.468 in cross-validation). SHapley Additive exPlanations (SHAP) were used to interpret the contributions of various predictors. They revealed that commuting patterns, such as the use of multiple modes of transportation, and employment in the health and social services sector were important factors associated with regional fertility levels. The results highlight the advantages of machine learning in capturing non-linear and interaction effects in fertility-related data, providing a foundation for informed policy-making. However, the ecological study design limits causal interpretations.