Abstract: The original F-score can only measure the discrimination of two sets of real numbers. This paper proposed the improved F-score which can not only measure the discrimination of two sets of real numbers, but also the discrimination of more than two sets of real numbers. The improved F-score and Support Vector Machines (SVM) were combined in this paper to accomplish the feature selection process where the improved F-score was used as the evaluation criterion of feature selection, and SVM to evaluate the features selected via the improved F-score. Experiments have been conducted on six different groups from UCI machine learning database. The experimental results show that the feature selection method, based on the improved F-score and SVM, has high classification accuracy and good generalization, and spends less training time than that of the Principle Component Analysis (PCA)+SVM method.

Key words: F-score, Support Vector Machine (SVM), feature selection, Principle Component Analysis （PCA）, Kernel Principal Component Analysis (KPCA)

摘要： 将传统F-score度量样本特征在两类之间的辨别能力进行推广，提出了改进的F-score，使其不但能够评价样本特征在两类之间的辨别能力，而且能够度量样本特征在多类之间的辨别能力大小。以改进的F-score作为特征选择准则，用支持向量机(SVM)评估所选特征子集的有效性，实现有效的特征选择。通过UCI机器学习数据库中六组数据集的实验测试，并与SVM、PCA+SVM方法进行比较，证明基于改进F-score与SVM的特征选择方法不仅提高了分类精度，并具有很好的泛化能力，且在训练时间上优于PCA+SVM方法。

关键词: F-score, 支持向量机, 特征选择, 主成分分析, 核函数主成分分析