一种适于心电数据的可分性评价准则研究

Research on a separability measurement criterion suitable for cardiac data

在心电特征提取中,现存的基于多种可分性准则和K L变换等技术在特征提取的有效性等方面都有各自的优缺点,都不能保证取得满意的结果。这就需要有一个评价准则来衡量最终特征的有效性和类间的可分性。为此,本文针对心电数据提出了一种基于标准差和欧氏中心距的可分性评价准则,并应用于MIT BIH数据库中心律失常数据的特征提取、特征有效性的评测和决策树的设计。实验结果表明,这是一种有效和实用的可分性评测准则。

Existing feature extraction techniques for cardiac signals are based on various separability criterion and like K-L transformation ect. Most of them have some advantages and disadvantages and are not always satisfactory for effective feature extraction. There is a need to develop a criterion to measure the performance of cardiac features and the separability between different classes. A separability measurement criterion suitable for cardiac data was presented, which was based on standard deviation and Euclidean center distance. The criterion was applied to cardiac arrhythmia data obtained from MIT-BIH database in order to extract features, measure the performance of features and build the decision tree for multiclass classification. The experimental results show that it is an effective and practical separability measurement criterion.