An Analysis of Methods for Tuning a Support-Vector Machine for Binary Classification

Abstract

Using methods based on support-vector machines (SVMs), it is possible to handle voluminous, high-dimensional, and poorly structured datasets, which is especially important in finding solutions for predictions in bioinformatics. In this paper, we discuss the key stage in designing support-vector machines, namely, how to choose the model. Methods for tuning support-vector machines were analyzed for binary classification. Two alternative approaches, that is, the gradient-based method and the derivative-free method of stochastic searching, were considered and performed for determination of the best values of hyperparameters. The quality performance of the support-vector classifiers obtained using the above-mentioned methods is investigated on the basis of benchmark data. The nested resampling technique is used to improve the accuracy of model evaluation. The results show the effectiveness of the chosen model-selection method for binary classification.