ニューラルネットによる判別モデルについて(一般講演4A)

抄録

In the present article, we propose a feed-forward neural network model for data of mutually exclusive groups and a set of predictor variables. We illustrate some applications of the bootstrapping which implement the bootstrap-based information criterion when selecting the optimum number of hidden units for a neural network model. We examine the thyroid disease database that is not only a pure training sample study which compares estimated measures of predictive performance, but test sample which reports the realized test sample error rates associated with a constructed prediction rule. In addition, influential analysis is also investigated based on deviance.