Handling Missing Data with the Tree-Structured Self-Organizing Map

In this paper we propose how a tree-structured self-organizing map (TS-SOM) can be used to impute incomplete data sets. The methodology has two parts, a new training algorithm utilizing incomplete data, and an imputation strategy that explains how the actual imputation is done. An introduction about evaluation studies of the proposed methodology is given also. Finally the performance of the methodology is demonstrated against standard methods using one simulated and one real world example.