Automatic probabilistic knowledge acquisition from data

Abstract

This paper documents an outline for a computer program for extracting significant correlations of attributes from masses of data. This information can then be used to develop a knowledge base for a probabilistic “expert system.” The method determines the “best” estimate of ~oint probabilities of attributes from data put into contingency table form. A major output from the program is a general formula for calculating any probability relation associated with the data. These probability relations can be utilized to form IFTHEN rules with associated probability, needed for expert system development.