Improving radial basis function kernel classification through incremental learning and automatic parameter selection

Training algorithms for radial basis function kernel classifiers (RBFKCs), such as the support vector machine (SVM), often produce computationally burdensome classifiers when large training sets are used. Furthermore, the developer cannot directly control this complexity. The proposed incremental asymmetric proximal support vector machine (IAPSVM) employs a greedy search across the training data to select the basis vectors of the classifier, and tunes parameters automatically using the simultaneous perturbation stochastic approximation (SPSA) after incremental additions are made. The resulting classifier accuracy, using an a priori chosen run-time complexity, compares favorably to SVMs of similar complexity whose parameters have been tuned manually.