An Investigative Approach to Employ Support Vector Classifier as a Potential Detector of Brain Cancer from MRI Dataset

Abstract

This paper proposes an investigative analysis to study the applicability of support vector classifier (SVC) algorithm to detect brain cancer efficiently. Brain cancer, mostly triggered by tumor cells in brain can be too lethal if malignancy is not identified at an early stage. Timely and tailored treatment plan will lead to an optimistic result which will lessen the magnitude of the disease. But it is really challenging to figure out malignancies manually from a large MRI dataset. In this context, a dataset from kaggle has been deployed to conduct multiclass classification by SVC where the information is extracted from MRI pictures. However, Linear SVC and NuSVC are also investigated apart from traditional SVC method. In order to increase the performance of the models, grid search cross validation is applied to tune the hyperparameters. All of confusion matrices for both tuning and without tuning of hyperparameters are presented in a comprehensive manner and thus, the performance parameters are tabulated and evaluated extensively. Among the three approaches, NuSVC depicts the maximum accuracy of 95.71% along with tuning the hyperparameters.