Output-coding and SVM for multiclass microarray classification

Multiclass cancer classification based on microarray data has been studied in this paper. A generalized output coding scheme combined with support vector machines as binary classifers; is used. Different coding strategies, decoding functions and feature selection methods are combined and validated on two cancer datasets: GCM and ALL. By using the random coding strategy and recursive feature elimination, the testing accuracy that we have achieved is as high as 80.4% on the GCM data which has 14 classes. Comparing with the other classification methods, our method has shown its superiority in classificatory performance.