Fast support vector classifier applied to microarray data

Since the early stages of the introduction of DNA microarray technology, there has been an enormous interest on clinical application for various diseases diagnosis. Microarray data classification is a difficult task for biologists due to its small sample sizes combined to its high number of features increasing the risk of overfitting. In the past years tools have been developed to extract biological information from microarray data but there is no common accepted method. In this paper we established a processing method based on a Fast Support Vectors Classifier and a feature selection scheme based on the R package LIMMA. The proposed method was tested on a lung cancer gene expression dataset provided as part of a competition called IMPROVER Diagnostic Signature Challenge. The scoring methods used to evaluate the algorithm performance were BCM, AUPR, CCEM as defined by IMPROVER organizers and results were encouraging.