Investigation of the effect of Training Data on the Performance of Support Vector Machine in Classification Brain of MR Images
International Journal of Emerging Trends in Science and Technology,
Vol. 1 No. 04 (2014),
2 June 2014
In recent years, a wide research is being carried out on brain imaging which involves computer aided detection of abnormalities in brain. Out of many diagnostic imaging techniques for the early detection of any abnormal changes in brain tissues, Magnetic Resonance Imaging (MRI) is a widely-used imaging method. The shortage of radiologists for analyzing the brain MR images calls for an automated system to analyze and classify such medical images. Support Vector Machine (SVM) has been widely used in the recent years to classify brain MR images into different classes. SVM Classifiers perform the task of classification in two phases – training phase and testing phase. The amount of image data to be used for training plays a vital role in determining the accuracy of the SVM. This paper focuses on determining the optimal number of image data in the training set for which a better classification accuracy is obtained. Classification experiments with various percentages of data in the training set show that 80% of total dataset is the optimal one. Results also point out that Polynomial kernel function of SVM is more apt for brain MR images classification with classification accuracy of 100% when trained with 80% of data.