Least squares support vector machine classifiers: an empirical evaluation

In this paper, we evaluate least squares support vector machine (LS-SVM) classifiers with RBF kernels on five publicly available real-life benchmark UCI data sets. While standard SVM optimisation involves solving quadratic or linear programming problems, the least squares version corresponds to solving a set of linear equations, due to equality constraints in the problem formulation of the SVM. Very promising results are reported indicating the good generalization behavior of the estimated RBF LS-SVM classifiers. For many large scale real life applications least squares support vector machines in combination with the tuning technique presented in this paper may offer a fast and simple method for obtaining classifiers with good generalization performance.status: publishe