Combining Mental Fit and Data Fit for Classification Rule Selection

Abstract

Mental fit of classification rules is lately introduced to judge the adequacyof such rules for human understanding. This paper first discusses the various criteria introduced in relation to mental fit in the literature. Based on this, the paper derives a general criterion for the interpretability of partitions generated by classification rules. We introduce interpretability as a combination of mental fit and data fit, or more specifically, as a combination of comprehensibility and reliability of a partition. Weintroduce so-called prototypes to improve comprehensibility, and the so-called reliability of such prototypes as a measure of data fit.