Abstract
This paper presents a general nonlinear approach to pattern classification using principal curves. Princi- pal curves(1) are non-parametric, nonlinear generaliza- tions of the first principal component, and may also be regarded as continuous versions of 1-D self-organizing maps (2), (3). The new classification technique, Princi- pal Curve Classifier (PCC), involves a novel way of com- puting a principal curve for each class using the class- labeled training data. An unlabeled test point is given the class-label of the principal curve that is closest to it in Euclidean distance. Preliminary experiments comparing the PCC with established classification methods, using selected datasets (originally from the UCI machine learn- ing repository(4)) from the Elena(5), (6) and Proben1(7) benchmarks , highlight the merits and limitations of this algorithm.

This publication has 16 references indexed in Scilit: