Event History Analysis Using Recurrent Neural Networks

Abstract

Recently, recurrent neural networks have been widely used in many fields. In the present study, we propose nonlinear analysis using recurrent neural network for leukemia disease data. Bone marrow transplants are a standard treatment for acute leukemia. Recovery following bone marrow transplantation is a multi-state process. We can select the optimum hidden unit based on bootstraping. Outliers are identified by using influential analysis. The significance of recurrent connection in recurrent neural networks is also tested. In order to summarize the measure of goodness-of-fit, the deviance on fitting of the recurrent neural network can be bootstrapped. This article examines predictions of probabilities at some points in multi-state survival models for processing a sequence of covariates values. By using recurrent neural networks, we can predict the conditional probability of surviving for the following short-term (say, six months) during the course of the disease with better accuracy than feed-forward neural networks, partial logistic models or Cox's proportional hazards model.