Empirical Comparisons of Neural Networks and Statistical Methods for Classification and Regression

Abstract

In this chapter we empirically compare neural networks and statistical methods on two telecommunications problems — one regression problem and one classification problem. Our goal is to explore the relative performance of neural networks and several statistical methodologies on moderate-sized regression and classification problems. A regression problem is one in which the goal is to estimate a numerical quantity based on partial input information; a classification problem is one in which the goal is to estimate the class to which an item belongs, again based on partial input information. We did not use a systematic method to choose problems for comparison, but rather chose one significant complex problem of each type. By significant, we mean that the underlying scientific or engineering question was important. By complex, we primarily mean that the dimensionality of the problem was large (roughly, greater than 20 measurement dimensions per observation or case).