Parameter analysis of multi class support vector machine for classifying rRNA from bacterial taxonomy

Support Vector Machines (SVMs), one of the new techniques for classifications, have been widely used in many application areas. And multiclass SVMs have been popular recently. The algorithm for SVMs is sensitive to the choice of parameter settings and the choice of these parameters can significantly affect generalization performance of SVM-classifiers. One of the parameters that has significant role on the performance of the SVMs is C, the regularization parameter. It is actually the trade-off between error and margin of the SVMs. The objective of this thesis is to analyze the impact of the parameter C and how to find an optimum way to select its value while training the SVMs for classifying Bacterial 16srRNA sequences. Cross validation has been proposed as a good way to find the optimum value of C for which the SVM can effectively classify the sequences with highest accuracy. Several experimental results have been reported to show the impact of C on the performance of multiclass SVMs.