A practical feature selection based on an optimal feature subset and its application for detecting lung nodules in chest radiographs

Abstract

The traditional motivation behind feature selection algorithms such as a genetic algorithm, a forward stepwise and a backward stepwise selections [1], is to find the best feature subset for a task using one particular learning algorithm. The idea is to select a optimal subset of attributes which are as representative as possible of the original data. However, it has been often found that no single classifier is entirely satisfactory for a particular task. Therefore, how to further improve the performance of these single systems on the basis of the previous optimal feature subset is a very important issue. Ensemble systems, also known as committees of classifiers, are composed of individual classifiers, organized in a parallel way and their outputs are combined in a combination method, which provides the final output of the system. Given the success of ensembles, ensembles allow us to get higher accuracy and sensitivity, which are often not achievable with single models. Based on the above, we propose a practical feature selection approach that is based on an optimal feature subset of a single CAD system, which is referred to as a multilevel optimal feature selection method (MOFS) in this paper. Through MOFS, we select the different optimal feature subsets in order to eliminate features that are redundant or irrelevant and obtain optimal features, and then a bagging ensemble with a MOFS method is proposed. Experimental results indicates that the accuracy of the bagging ensemble using a MOFS method is superior to that of a single CAD system and is also superior to that of the ensemble using an attribute selection algorithm based on ReliefF.