Using Artificial Neural Nets for Statistical Discovery: Observations after Using Backpropogation, Expert Systems, and Multiple-Linear Regression on Clinical Trial Data
Erach A. Irani
John P. Matts
John M. Long
James R. Slagle
POSCH group
University of Minnesota, Minneapolis, MN 55455, USA
Abstract
Powerful new training algorithms developed for artificial neural networks hold the promise that they can identify regularities in the training data and generalize over the test data. The backpropagation algorithm is one such training algorithm that, with the use of hidden units, can learn functions such as exclusive-or. These functions can be learned by statistical techniques such as multiple-linear regression only by introducing additional parameters. We report experimental comparisons of the performance of backpropagation, multiple-linear regression, and an expert system. We conclude that, for the data studied here, backpropagation is unsuitable for discovering statistical relationships. It may be possible to customize neural-net algorithms for niche applications in discovery systems.
