Support vector machines

Support vector machines (SVMs), ¹ originally developed by Vladimir Vapnik and his coworkers (e.g., Boser et al., 1992; Guyon et al., 1993; Cortes and Vapnik, 1995; Schölkopf et al., 1995, 1996; Vapnik and Lerner, 1963; Vapnik and Chervonenkis, 1964, 1979; Vapnik et al., 1997), and belonging to the class of supervised learning (they go through statistical learning ² ) can be used for optical character recognition (OCR); pattern recognition such as handwriting, speech, images, etc.; classification and regression analysis; and time series prediction. The simplest form of SVMs is when a given set of input data has to be classified into two possible classes, thereby leading to binary classification. Mathematically, SVMs achieve the classification by constructing a hyperplane in a high-dimensional space. By aiming at a larger separation between the hyperplane and the nearest data point, a lower generalization error can be expected. SVMs have good generalization properties, but are rather slow in the testing phase. They are universal approximators just as multilayer perceptrons and radial basis functions (RBFs).