An Imputation Method for Missing Data Based on an Extreme Learning Machine Auto-Encoder

Abstract

This paper proposes an imputation method for missing data based on an extreme learning machine auto-encoder (ELM-AE). The imputation chooses a set of plausible values determined by ELM-AE and then substitutes the average of these plausible values for the missing values. To compare the performance of ELM-AE imputation with the three other widely used imputation techniques, we conducted comprehensive experiments using seven UCI benchmark data sets. The proposed ELM-AE imputation approach proved to be superior to the other three methods based on the results using these data sets.