Learning Symbolic Prototypes

Abstract

We present an empirical analysis of symbolic prototype learners for synthetic and real domains. The prototypes are learned by modifying the minimum-distance classifier to solve problems with symbolic attributes, attribute weighting, and its inability to learn multiple prototypes for a class. These extensions are implemented in SNMC. In the second half of this paper, we provide empirical analysis, characterizing situations where symbolic prototypes have advantages over traditional methods such as decision trees and instance-based methods. Empirical analysis on real domains show that SNMC increases classification accuracy by 10% over the original minimum-distance classifier and has a higher average generalization accuracy than both C4.5 and PEBLS on 20 domains from the UCI data repository. Finding multiple prototypes for classes results in the same or higher accuracy than learning a single prototype for classes. Keywords: Inductive learning, classification, instance-based learning, proto...