Construction of diagnostic rules based on assessments of subjects

This paper is based on assessments of subjects which are not defined by formal procedures. Logically structured assignment rules are constructed which are stable over time and which have a low number of disagreements with the assessments of the subjects. Recent results obtained from a computational learning theory show that the proposed construction of rules leads with high probability to rules with similar performances on large enough learning and validation sets of subjects. The construction of a set of stable rules is illustrated by using as a case study the psychophysiological, personality and speech behaviour data of a repeated-measures Type A behaviour study. This case study is also used for a discussion of the issue of biased ratings in the presence of assessments which are not defined by formal procedures.