Optimal Feature Subset Selection Method for Improving Classification Accuracy of Medical Datasets

Abstract

            Medicine is indeed one of the sciences in which computer science advancement makes a great deal of progress. The use of medical computers enhances precision and speeds data processing and diagnosis processes. There are now various diagnostic systems assisted by computers and machine-learning algorithms play a key role. More precise and faster systems are needed. The classification is the popular machine-learning job, part of computer-aided diagnostic systems and various packages of medical data analysis software. It is necessary to choose functional set and proper parameters for the classification model to achieve higher classification accuracy. Medical databases also contain wide set of features where many features correlate with others, so reducing the set of features is essential. Most classifiers are structured so that they can learn from the data themselves through a training process because full expert experience is not realistic to evaluate classification parameters.In this paper, in order to improve the accuracy of the classifier with proposed Feature Selection method. Differential evolution optimization is used to find the optimal subset obtainedby the Filter based feature selection method.The performance of the proposed feature selection is evaluated with classifiers like Random Forest Classifier, Gradient Boosting Tree, Artificial Neural Network and Support Vector Machine.