A Hybrid Method for Random Pattern Sequence Classification

Abstract

We present a hybrid method for random pattern sequence classification that takes into account the random structural properties of the sequence. The method works in two steps. A segmentation step, dividing the original sequence into segments, such that all observations in a same segment belong to a unique class, and a classification step, where each segment is classified by a neural network classifier.

Keywords:

• Classification problem
• Neural networks
• Pattern classification
• Simulated annealing
• Segmentation

Acknowledgments

Supported by Universidad Simón Bolivar, Decanato de Investigación y Desarrollo de la USB, CDCH Universidad Central de Venezuela grant and ECOS-NORD project.