Red neural multitarea para problemas de decisión con información incompleta

Abstract

Missing data is a common problem that appears in many real applications. Handling missing data is a essential requirement for pattern classification because inappropriate treatment of missing data may cause large errors or false results on classification. A classical approach is to estimate and fill the missing values. Up to now, proposed methods are efficient but they do not focus the missing data estimation to pattern classification. In this work, we propose a novel neural network that uses the advantages of Multitask Learning (MTL) to classify patterns with incomplete data. A MTL neural network learns at the same time the classification task and the different task associated to incomplete features. Missing data estimation is guided and oriented by the classification task during the MTL process. Obtained results based on real and artificial classification problems demonstrate the advantages of the proposed algorithm.