Flexible control of case-based prediction in the framework of possibility theory (EWCBR 2000)
Résumé
The "similar problem-similar solution" hypothesis underlying case-based reasoning is modelled in the framework of possibility theory and fuzzy sets. Thus, case-based prediction can be realized in the form of fuzzy set-based approximate reasoning. The inference process makes use of fuzzy rules. It is controlled by means of modifier functions actingo n such rules and related similarity measures. Our approach also allows for the incorporation of domain-specific (expert) knowledge concerning the typicality (or exceptionality) of the cases at hand. It thus favors a view of case-based reasoning accordingto which the user interacts closely with the system in order to control the generalization beyond observed data. Our method is compared to instance-based learning and kernel-based density estimation. Loosely speaking, it adopts basic principles of these approaches and supplements them with the capability of combining knowledge and data in a flexible way.