IEEE ICHI Data Analytics Challenge on Missing data Imputation by Amelia II

Abstract

There are some traditional tools for the imputation of missing data, such as imputeTS [1], multiple imputation with chained equations (MICE) [2] and Gaussian process (GP) [3]. The imputeTS package and GP specializes on univariate time series imputation, MICE could not take the advantage of time information. 3D-MICE [4] algorithm combining MICE-based with GP-based predictions to impute missing data based on both cross-sectional and longitudinal information. The reason we chose Amelia II [5] was that it could utilize both time series and multivariable information at one model, implying that Amelia II may be a more suitable tool for MIMIC dataset. Amelia II uses the bootstrap-based EMB algorithm to impute many variables with many observations. EMB algorithm combines the classic EM algorithm with a bootstrap approach.