Predicting the growth and key influencing factors of home-visit nursing offices in Japan using machine learning

Ishikawa, Takemasa (2025) Predicting the growth and key influencing factors of home-visit nursing offices in Japan using machine learning. World Journal of Advanced Research and Reviews, 25 (2). pp. 2284-2294. ISSN 2581-9615

The distribution of home-visit nursing offices in Japan is uneven, with some municipalities facing shortages. Understanding the factors influencing their growth rate is crucial for policy planning. This study developed a machine learning model to predict the growth rate of home-visit nursing offices using municipality-level time-series data from 2015 to 2022. Demographic indicators, healthcare resources, and economic factors were incorporated as predictors. Extreme Gradient Boosting (XGBoost) was employed, integrating one-year and three-year lag variables and a three-year moving average to capture temporal trends. Model performance was assessed using R², and Shapley Additive Explanations (SHAP) analysis was conducted to interpret feature importance. The model demonstrated strong predictive performance, with an average R² of 0.87. The past number of home-visit nursing offices had the highest impact on growth, with the three-year moving average contributing positively and the one-year lag variable indicating potential market saturation. Population density was also positively associated with growth. Although the aging rate had a limited overall influence, a higher aging rate tended to be associated with a lower growth rate of home-visit nursing offices. Economic indicators and the number of hospitals had minor influences. These findings suggest that market conditions and supply-side constraints significantly shape the expansion of home-visit nursing offices. Strategic interventions, such as financial support in underserved areas and sustainability measures in saturated regions, are needed to ensure an optimal distribution of services. Future research should explore additional socioeconomic factors and external shocks to refine predictive models and support data-driven policymaking.