SVM Model Selection for Microarray Classification

Support vector machines (SVMs) [4] are gaining broad acceptance as state-of-the-art classifiers for microarray data analysis [3]. However, most studies that use SVMs to predict sample class consider only a small subset of SVM kernels and parameters. The effect of the kernel type and parameter values is usually not studied in microarray classification. The choice of kernel and classifier parameters is a form of model selection. Although the machine learning community has extensively considered model selection with SVMs [2], optimal model parameters are generally domain-specific. The present study evaluates the impact of kernel type and parameter values on the accuracy with which a SVM can classify microarray data. We hypothesized that classification accuracy would vary with the kernel type and parameter values, and that the optimal parameter values would vary with the kernel type.