[2J1-GS-2] Machine learning: Gaussian process model

[2J1-GS-2-04] Data fusion method with Gaussian process in NMAR missingness

〇Masaki Mitsuhiro^1,2, Takahiro Hoshino^3,4 (1. Nikkei Research Inc., 2. Graduate School of Economics, Keio University, 3. Department of Economics, Keio University, 4. RIKEN Center for Advanced Intelligence Project)

Keywords:Statistical data fusion, Gaussian process latent variable model, Missing data

It is necessary to integrate multiple-source datasets into a single-source dataset to investigate the relationship between variables that are not observed simultaneously.
Among various data fusion methods, latent variable modeling, which assumes common factors behind covariates and outcome variables, cannot capture the non-linear relationship between observed and latent variables.
In the field of machine learing, the Gaussian process latent variable model, which is an extension of the principal component analysis, has been proposed and it is good when extracting features of the non-linear relationship between variables.
In this study, we propose a data fusion method with Gaussian process.
This proposed method can be apply to NMAR missing data and we can deal with discrete variables as well as continuous variables.