A Genetic Programming Based ECOC Algorithm for Microarray Data Classification

Microarray technology aims to discover the relationship between genes and cancers. But the analysis of multiclass microarray datasets is a difficult problem in considering the small sample size along with the class imbalance problem. In this paper, we propose a Genetic Programing (GP) based Error Correcting Output Codes (ECOC) algorithm to tackle this problem. In our GP framework, each individual represents a codematrix, and a legality checking mechanism is applied to avoid the production of illegal codematrices. So the algorithm evolves towards optimum ECOC codematrices. In experiments, our algorithm is compared with other methods based on four famous microarray datasets. Experimental results prove that our algorithm can achieve better results in most cases.