Finding Dominant Factor That Affects Crude Birth Rates in Japanese Prefectures

Abstract

We conduct a regression to find a dominant factor that affects crude birth rates in Japan by prefectures. As the traditional regression method, a linear multiple regression is widely used. However, higher accuracy methods with machine learning algorithms have been developed. To find the dominant factor, we use eXtreme Gradient Boosting (XGBoost) and Random Forest which are the decision tree based machine learning algorithms. The results show better accuracies, compared with the traditional linear multiple one. Then, the XGBoost shows that the most dominant factor is the number of marriages, and the second one is the migration rate to the prefecture.