Effort to Develop a General Methodology and Tool for Prediction Error Decomposition

The Data Science Related Basic Research Working Group of the Institute of Actuaries of Japan (IAJ) is making an effort to develop a general methodology and tool for decomposing prediction errors into process errors, parameter errors, and other errors, which is widely applicable to varieties of predictive modeling methods, even to the machine learning methods. In this presentation, we will explain the current status of development and demonstrate what can be done by our research outcome by using numerical examples.