Subject specific treatment to neural networks for repeated measures analysis

Abstract

Analysis of repeated measures data for the purpose of prediction is not an easy task particularly when the problem under consideration is highly nonlinear, number of subjects is large and the sample available to learn the model is small. The efficacy of the ANN for subject level treatment has been studied here empirically. Data were generated through a random coefficient model and a few nonlinear mixed effect models. For ANN feedforward backprop has been tried. Simulations have been conducted with varying number of covariates and parameters (both common and subject dependent), number of subjects and different sizes of repeated measures. ANN has demonstrated considerable promise.