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

Several recent machine learning publicationsdemonstrate the utility of using feature selectionalgorithms in many learning. Feature selection helps toacquire better understanding about the data by tellingwhich the important features are and how they arerelated with each other and it can be applied to bothsupervised and unsupervised learning. This paper aimsto find the best subset of features that not onlymaximizes the classification accuracy but minimizes thenumber of features. The other reason is to make awareof the necessity and benefits of applying featureselection methods. In this paper, genetic algorithm isone of the wrapper feature selection methods and it isused to reduce the irrelevant attributes of data.Embedded feature selection method (C4.5) is used toprune the features selected by genetic algorithm whichis suffering from overfitting problem. By combininggenetic algorithm with decision tree, this methodenhances the Bayesian classification to eliminateunnecessary features and produces accurate classifier