VARIABLE SELECTION AND NEURAL NETWORKS APPLIED TO THE CLASSIFICATION OF RISK OF ADVERSE EVENT IN CHILDHOOD LEUKEMIA

摘要

In this paper we look at an application of input variable selection methods, in connection with a Feedforward Neural Network, with the goal of evaluating the risk of adverse event (relapse or death in complete remission) in a group of children with Acute Lymphoblastic Leukemia. This is an important medical problem where the number of available input patterns is always quite small, making an appropriate choice of inputs, crucial. The obtained performance showed excellent results for generalization (leave-one-out approach), reaching a rate of more than 96% of correctness for all simulations.