1. School of Computer Science, Shaanxi Normal University, Xi’an Shaanxi 710062, China 2. School of Electronic Engineering, Xidian University, Xi’an Shaanxi 710071, China 3. School of Information Engineering, Shenzhen University, Shenzhen Guangdong 518060, China
Abstract：As a criterion of feature selection, F-score does not consider the influence of the different measuring dimensions on the importance of different features. To evaluate the discrimination of features between classes, a new criterion called D-score was presented. This D-score criterion not only has the property as the improved F-score in measuring the discrimination between more than two sets of real numbers, but also is not influenced by different measurement units for features when measuring their discriminability. D-score was used as a criterion to measure the importance of a feature, and Sequential Forward Search (SFS) strategy, Sequential Forward Floating Search (SFFS) strategy, and Sequential Backward Floating Search (SBFS) strategy were, respectively, adopted to select features, while Support Vector Machine (SVM) was used as the classification tool, so that three new hybrid feature selection methods were proposed. The three new hybrid feature selection methods combined the advantages of Filter methods and Wrapper methods where SVM played the role to evaluate the classification capacity of the selected subset of features via the classification accuracy, and leaded the feature selection procedure. These three new hybrid feature selection methods were tested on nine datasets from UCI machine learning repository and compared with the corresponding algorithms with F-score as criterion of the discriminability of features. The experimental results show that D-score outperforms F-score in evaluating the discrimination of features, and can be used to implement the dimension reduction without compromising the classification capacity of datasets.
2011, Vol. 31 
