Improving the imbalance between supply and demand for nursery schools is an important issue in recent years. This paper aims to build a machine-learning model that estimates which nursery school each household will choose. We train the model, LightGBM, using (i) detailed information on the actual choice of nursery schools by each household collected through a Web questionnaire (conducted on 1,032 households across Japan) and (ii) attribute information of nursery schools all over Japan gathered from open data. The F-score for the trained model was 68.7％. Furthermore, based on the SHAP values calculated using the trained model, we demonstrate the quantitative impact of each explanatory variable (such as household attributes, facility characteristics, and commuting conditions) on the household nursery school choice.