Neural Regression Model, Resampling and Diagnosis

摘要

In this article logistic regression model is highlighted using a feedforward neural network model. Suggested here is a simple more precise prognostic technique by incorporating follow-up data in the development of the model when the data consist of grouped binary response and a set of predictor variables, which is closely related to classical logistic regression model. Statistical techniques are formulated in terms of the principle of the likelihood when a resampling process such as bootstrapping and cross-validation is applied. We then attempt to determine the number of units in the hidden layer, to verify the asymptotic χ{sup}2 behavior of the deviance for the goodness-of-fit test of the model, to detect inappropriate influential observations and outliers, and to improve the goodness-of-fit test of the neural network model.