A Comparison of Multi Support Vector Machine Performance with Popular Decomposition Strategies on Alzheimer’s Data

Abstract

In Medical field, classifying the images into more than two diagnostic classes is still remaining as a challenge. Particularly in diagnosis of Alzheimer’s the brain MRI images should be classified into more classes for effective diagnosis. Multi Support Vector Machine (MSVM) is advancement to the standard SVM which can able to deal multi-class classification problem efficiently. MSVM is successfully applied in the fields of Text categorization, Medical Image classification, Handwriting Recognition, Protein Structure Prediction, etc. MSVM uses numerous approaches such as Directed Acyclic Graph, One-vs-One and One-vs-All etc., to classify the data into multi classes. Among them “One vs One” and the “One vs All” are the important decomposition strategies. Hence, the both decomposition strategies are applied separately with MSVM in this study to classify the Alzheimer’s disease data. The performance of MSVM with these two strategies is compared and the best decomposition strategy is identified.