Forecasting Japanese Elections by Applying Ensemble Learning Methods

TKFD Working Paper Series No. 20-05


This paper applies machine learning methods, namely, the decision tree algorithm and ensemble learning methods to forecast Japanese lower house elections. By applying these two machine learning methods, we developed several non-linear forecasting models. We then compared our forecasting results with those generated by Lewis-Beck and Tien (2012), a pioneering model on this topic, using the same data and the same explanatory variables. We found that, even without tuning, all of our ensemble-learning-based forecasting models exceeded Lewis-Beck and Tien’s model in mean accuracy. All of them also provided better explanations for mean variance. Despite small sample size inherent for country-specific forecasting models, our non-linear forecasting models of Japanese election generated better performance than linear models. We believe, by combining with substantive theories of each country’s political situation, our methodological approach can improve predictability of country-specific forecasting models of other countries.

Download Working Paper (PDF: 656 KB)

阿山 晴取/Budrul Ahsan

Budrul Ahsan


Areas of Expertise

  • Deep learning
  • data mining
  • recommendation system
  • genome analysis

Research Unit

Politics & Economy

加藤 創太/Sota Kato

Sota Kato


Areas of Expertise

  • Comparative political economy
  • political methodology
  • public opinion

Research Unit

Politics & Economy

茨木 瞬/Shun Ibaragi

Shun Ibaragi