An Efficient Sequential RBF Network for Gene Expression-Based Multi-category classification

Abstract

This paper presents a fast and efficient sequential learning method for RBF networks that can perform classification directly for multi-category cancer diagnosis problems based on microarray data. The recently developed algorithm, referred to as Fast Growing And Pruning-RBF (FGAP-RBF) can perform incremental learning on the future data directly. No training of all the previous data is needed. This character can reduce the learning complexity and improve the learning efficiency and is greatly favored in the real implementation of a gene expression-based cancer diagnosis system. We have evaluated FGAP-RBF algorithm on a benchmark multi-category cancer diagnosis problem based on microarray data, namely GCM dataset. The results indicate that compared with the results available in literature FGAP-RBF algorithm produces a higher classification accuracy with reduced training time and implementation complexity.