Feature selection for classification modeling has been attracting increasing attention in many industries particularly in big data processing for its advantages in improving the predictive efficiency, enhancing the intelligibility and reducing the cost of feature acquisition.Different from extant research, we regard feature selection in this paper as an efficiency evaluation process with multiple inputs and outputs and propose a novel feature selection framework based on Data Envelopment Analysis (DEA).We then propose a simple feature selection method based on the framework, where the inputs and outputs make the method supervised learning oriented.Experimental results on twelve UCI datasets indicate that proposed method is effective and outperforms several representative feature selection methods in most cases.The results also show the feasibility of proposed DEA-based feature selection framework.