An optimal classification rule for multiple interval-screened scale mixture of normal populations

This paper considers a new classification case where K categories of the target variable are defined by disjoint intervals of an underlying variable and proposes an optimal rule for the classification. A parametric classification model, known as interval-screened scale mixture of normal model, is used to derive the rule that classifies individuals into K populations defined by K disjoint intervals of the variable (screening variable). The effectiveness of the rule is verified by the simulation and empirical studies that compare its performance with other existing classification rules. The cross-validation error rate is used as the measure of performance. Necessary theories for deriving the rule, an MCEM algorithm for estimating the rule, and the interesting characteristics of the rule are also provided.