Fast missing value imputation using ensemble of SOMs

Abstract

This report presents a methodology for missing value imputation. The methodology is based on an ensemble of Self-Organizing Maps (SOM), which is weighted using Nonnegative Least Squares algorithm. Instead of a need for lengthy validation procedure as when using single SOMs, the ensemble proceeds straight into final model building. Therefore, the methodology has very low computational time while retaining the accuracy. The performance is compared to other state-of-the-art methodologies using two real world databases from different fields.