Hybrid learning of wrapper and embedded method for feature selection of medical data

Abstract

Several recent machine learning publications demonstrate the utility of using feature selection algorithms in many learning. Feature selection helps to acquire better understanding about the data by telling which the important features are and how they are related with each other and it can be applied to both supervised and unsupervised learning. This paper aims to find the best subset of features that not only maximizes the classification accuracy but minimizes the number of features. The other reason is to make aware of the necessity and benefits of applying feature selection methods. In this paper, genetic algorithm is one of the wrapper feature selection methods and it is used to reduce the irrelevant attributes of data. Embedded feature selection method (C4.5) is used to prune the features selected by genetic algorithm which is suffering from overfitting problem. By combining genetic algorithm with decision tree, this method enhances the Bayesian classification to eliminate unnecessary features and produces accurate classifier.