Competence reasoning — handling ambiguous and imprecise data in the Pro.M.D. expert system shell

Abstract

The Pro.M.D. expert system shell (Pohl and Trendelenburg, Methods Inf Med 1988;27: 111–117) is a computer program tool for knowledge-based computer interpretation of laboratory data. Competence reasoning deals with more or lessvague knowledge. Vague knowledge is knowledge using vague data. In Pro.M.D. terminology, vague data are called ‘semiknown’. Examples of semiknown numbers are numeric intervals like [5.2, 8.8] (= all numbers between lower limit 5.2 and upper limit 8.8 are possible), or imprecise numbers like N[4,1] (= numbers vary around mean of 4 with standard deviation of 1). Semiknown qualitative data are ambiguous information on alternative values like low or normal (= from three alternatives low or normal are possible and high is impossible) or probability distributions on the alternatives like low in 60% or normal in 30% or high in 10% (= low, normal and high appear with the respective probabilities). Regarding vagueness, data may be divided into known data (unambiguous and precise information), semiknown data (ambiguous or imprecise information) or unknown data (no information). Semiknown data may be further subdivided into probabilistic data, where the probability distribution of domain values is known (e.g. Gaussian distribution) and possibilistic data, where probability distribution of domain values is not known (e.g. intervals). Probabilistic data may be obtained either by completely counting the population's individuals or by estimation from samples. In the latter case sample size is another measure for data vagueness. In the Pro.M.D. expert system shell functions are being implemented for input, output, calculation and testing of semiknown data. Thus, instruments for competence reasoning in Pro.M.D. will help to build up reliable and user friendly knowledge-based systems.