Analysis of the MSVMO 0 ethod - the 3 robabilistic SVM 2 utputs
Ana Madevska-Bogdanova, Žaneta Popeska, Dragan Nikolik · 2005
Number of experiments has shown that for classification problems, if SVM models trained over the same data set have similar SVM outputs with respect to a given input vector x, the values of the corresponding MSVMO probabilities (3) have small differences. This means that the information from the MSVMO method from different SVM models trained over the same data set is essentially the same.