Validation of a hybrid approach for imputing missing data

A hybrid system has been constructed to impute missing values in a neonatal intensive care unit database using artificial neural networks and case-based reasoning. This paper presents the preliminary test results of a system using the connection weights of a linear neural network as the match weights in a case-based reasoner to find the closest-matching cases. The means of the ten closest-matching cases then replaced the missing values in the queries. The hybrid approaches were compared to mean and random imputations, and showed slightly better performance.