Optimal Deep Transfer Learning Framework for Feature Selection based Brain Tumor Image Classification

Abstract

Brain tumor(BT) is an important disease which affects major part of the global population including children and older people. The possibility of survival rate can be improved by the identification of the tumour at the starting level. The aspiration of feature selection (FS) method is to choose a smaller subset of features which will reduce the redundant part and increase the related part towards the target like class labels. Therefore, machine learning approaches gains maximum performance while including the chosen important features. So, presently, FS acts an important role in the classification task. Keeping this in mind, in this paper, we introduce a feature selection-based BT classification model. For FS, modified genetic algorithm (MGA) is used and deep transfer learning model is used for classification. To develop MGA, appropriate changes have been made in the traditional GA by minimizing the arbitrary behaviour. The presented DTL with MGA is validated on the collection of benchmark images and verified that the DTL results are improved by the use of FS methods.