Unsupervised conceptual learning in a diagnostic expert system

A diagnostic expert system for assessing colonic sections is augmented by the addition of an unsupervised learning module. A simple distance metric between diagnostic clue sequences, as they occur for each assessed case, is defined. The statistical significance of the modes detected in the conceptual data sets is established. Even the relatively simple unsupervised learning module implemented in this system has led to a number of insights. For example, if learning capability is considered for a diagnostic expert system, it is advisable to use a fine grading and multiple diagnostic clue values. This will allow better resolution by a distance measure. Also, it is possible to establish distances between concepts if an ordering can be attained and, based on such an ordering, the statistical significance of a grouping of cases, described only in conceptual terms, can be established