An Improved Principle for Measuring Generalization Performance
Zhou Wei · Chinese Journal of Computers · 2003
A new constructive principle, which depends on the distribution of examples, for measuring the generalization performance is proposed based on the analysis of the generalization performance of support vector machines. The principle is consistent in geometry with that in statistical learning theory, composed of two-order statistic of samples and shows the convergence rate of learning process well. It is important that this new principle can be processed before learning. So this new principle can be taken as a rule for all classifiers to preprocess data and to select model for SVMs.Simulation results for both artificial and real data show the rationality of this principle.