Trading off Coverage for Accuracy in Forecasts: Applications to Clinical Data Analysis

Artificial Intelligence in Medicine: Interpreting Clinical Data

Volume

Papers from the 1994 AAAI Spring Symposium

Contents

In this paper, we will explore how to modify a variety of machine learning algorithms to trade off coverage for accuracy. The key to these modifications is that learning components of many algorithms already have some internal measure of the quality of their concept description. These measures are typically used by the learner to favor one hypothesis over another, or to decide whether to prune a hypothesis. By giving the classification component access to these measures, we can allow the classifier to decide to classify only those examples for which its prediction is more likely to be correct. Here, we will show how the algorithms can be modified to trade off coverage for accuracy.