Variations of Support Vector Machine classification Technique: A survey

The Support Vector Machine (SVM) technique is emerged as a machine learning method used for classification, highly efficient and effective in the field of various applications like pattern recognition, image processing, fraud detection, text categorization etc. Its accuracy, robustness and providing best classification function to distinguish between members of the two classes in the training data are the main advantages, but the disadvantages can’t be ignored even. The memory requirement and computation complexity are the main disadvantage of it. Many techniques are developed to overcome these limitations which are broadly classified into decomposition based and variant based algorithms. Also, SVMs were originally developed to perform binary classification. However, applications of binary classification are very limited. Most of the classification problems involve more than two classes. A number of methods to generate multiclass SVMs from binary SVMs have been proposed and is still a continuing research topic. In this paper, we present the survey of such techniques and falls them into three groups. The decomposition based method: overcome memory limitation, variant based techniques: reduce the computational complexity, and multiclass based methods handle the multi class classification.