Article Preview
Top1. Introduction
Classification of data (Aggarwal 2014), (Duda, Hart & Stork, 2007) is a common task in machine learning. In this direction support vector machines (SVM) (Burges, 1998) has emerged as a promising pattern classification tool in recent years. It is based on the principle of structural risk minimization (SRM) and statistical learning theory (Vapnik, 1998). SVM was proposed by Vapnik (Vapnik, 1998) and has received much attention from the pattern recognition community (Abe, 2010), (Bishop, 2006), (Duda, Hart & Stork, 2007). It has been widely used in various real life applications with appreciable classification performances (Burges, 1998).
Many complex problems have been solved by SVMs (Abe, 2010). Some notable applications where SVM has been successfully applied are handwritten digit recognition, object recognition, speaker identification, charmed quark detection, face detection, optical character recognition, medical diagnostics, text classification etc. (Abe, 2010). Two important applications where SVM has outperformed other methods are electric load prediction (EUNITE, 2001) and optical character recognition (Tautu & Leon, 2012). For regression estimation SVMs have been compared on benchmark time series prediction tests, the Boston housing problem and (on artificial data) on PET operator inversion problem (Abe, 2010), (Burges, 1998). In most of these cases SVM generalization performance i.e. error rates on test sets either matches or is significantly better than that of the competing methods. The use of SVMs for density estimation and ANOVA decomposition has also been studied (Burges, 1998). Regarding extensions the basic SVMs contain no prior knowledge of the problem. For example, a large class of SVMs for image recognition problem gives the same results if pixels are first permuted randomly with each image suffering the same permutation, an act of vandalism that would leave the best performing neural networks severely handicapped. Although SVMs have good generalization performance they can be abysmally slow in test phase a problem which has been addressed in (Burges, 1998). Several works have generalized the basic ideas of SVM and have shown connections to regularization theory (Abe, 2010), (Burges, 1998). They have also shown how SVM ideas can be incorporated in a wide range of other algorithms (Abe, 2010), (Burges, 1998), (Chaudhuri, De & Chatterjee, 2008), (Chaudhuri & De, 2011), (Chaudhuri, 2014).