Diagnosis. II. Diagnostic models based on attribute clusters: A proposal and comparisons

Abstract

A new discrimination procedure based on the formation of clusters of dependent attributes, and estimation of the joint probability distribution as the product of the probabilities of the disjoint clusters is proposed and investigated. The major advantages of this method are a substantial reduction of the number of probability estimates that must be made, the ability to include symptom dependencies, and the ease and flexibility of its implementation.Comparisons with other discrimination procedures are obtained using Monte Carlo techniques. Results indicate that the proposed model is robust and may lead to gains over the independence and actuarial models, especially for small sample sizes.