MULTIVARIATE RULE BUILDING EXPERT SYSTEM

A multivariate rule building expert system (MuRES) has been devised that acquires knowledge In the form of rules from training sets of multivariate analytical data. MuRES differs from conventional expert systems In that It generates rules that consist of a linear combination of all variables. MuRES advantageously combines the robust character of univariate expert systems and the feature transformation properties of linear classifiers. MuRES develops its own certainty factors and provides qualitative information regarding the classification. When evaluated with exemplary data, MuRES performed better than linear discriminant analysis, soft Independent modeling of class analogy, and a univariate expert system. © 1990, American Chemical Society. All rights reserved.