On a stacking model among models with different target variables

Abstract

Developing a highly accurate prediction model for decision making is an important issue for companies.
However, as objective variables are set according to real business requirements, their structure often becomes complicated.
As one method of improving accuracy in such cases, there is a framework of ensemble learning that uses multiple models.
In particular, the method of using the prediction value of the model of the first stage as input of the next model is called stacking.

In this research, we propose a new stacking model which has a simple structure first stage model and a complicated structure next stage model, and we introduce examples of modeling using actual data of the automobile industry.