Blood transfusion prediction using restricted Boltzmann machines

Abstract

The availability of blood transfusion has been a recurrent concern for medical institutions and patients. Efficient management of this resource represents an important challenge for many hospitals. Likewise, rapid reaction during transfusion decisions and planning is a critical factor to maximize patient care. This paper proposes a novel strategy for predicting the blood transfusion need, based on available information, by means of Restricted Boltzmann Machines (RBM). By extracting and analyzing high-level features from 4831 patient records, RBM can deal with complex patterns recognition, helping supervised classifiers in the task of automatic identification of blood transfusion requirements. Results show that a successfully classification is obtained (96.85%), based only on available information from the patient records.