A Framework for Multi-Class Learning in Micro-array Data Analysis

A large pool of techniques have already been developed for analyzing micro-array datasets but less attention has been paid on multi-class classification problems. In this context, selecting features and quantify classifiers may be hard since only few training examples are available in each single class. This paper demonstrates a framework for multi-class learning that considers learning a classifier within each class independently and grouping all relevant features in a single dataset. Next step, that dataset is presented as input to a classification algorithm that learns a global classifier across the classes. We analyze two micro-array datasets using the proposed framework. Results demonstrate that our approach is capable of identifying a small number of influential genes within each class while the global classifier across the classes performs better than existing multi-class learning methods.