Summary Data Analysis and Expert Systems: Generating Rules from Data
Mireille Gettler-Summa, Henri Ralambondrainy, Edwin Diday · 1988
The combined viewpoints of Data Analysis and Artificial Intelligence are helpful in approaching certain problems with reviewed efficiency; for example, an easily compre hensible description of clusters may be provided by a conjunctive representation of variables in clustering; and in the techniques of learning from examples, a reduction of the order of an algorithm complexity may be obtained by hierarchical methods. In order to work in an area dealing with data, new tools of formalization are needed. For example, these symbolic data demand an extension of classical rectan gular array of Data Analysis: case may contain an interval or several values instead of a single value as usual, and objects are not necessarily defined by the same variables. Three algorithms are based on that knowledge representation. The first one gen erates clustering characterized by conjunctions of events. The second and the third generate that can be used in the knowledge base of an expert system: they use logical operators and counting procedures simultaneously. The purpose is to produce defining membership for objects belonging to a class of examples, and for the class of counterexamples, with discriminant and covering criteria. One of these al gorithms requires the non overlapping rules constraint; the other does not have that constraint and its complexity may be improved by a data analysis method. They are presented with computer programs and examples on marketing data sets.