Performance comparison issues in neural network experiments for classification problems

Considers the methodological aspects of neural network experiments in business applications. They emphasize the need for a more statistically rigorous comparison of neural nets with other traditional techniques. Specifically they identify several measures for estimating the performance of a classification technique. They illustrate these ideas through a comparison of neural nets and discriminant analysis. The results show that a much better picture of the performance capabilities of a technique emerges as a result of this additional analysis.<
>