常用分类算法在脑梗死预测中的比较研究

Comparison Study of Prediction of Cerebral Infarction Cases Based on Common Classification Algorithms

目的:比较常用分类算法对脑梗死的分类预测能力。方法:将反映动脉弹性的6个脉搏波参数加年龄、性别一共8个指标作为每个样本的特征。把样本按3∶1随机分为训练集和测试集两部分。分别利用人工神经网络(ANN)、贝叶斯(Bayes)、决策树(Decision Tree,DT)、K邻近法(k-NN)、支持向量机(SVM)算法构造分类器,使用各分类器对训练集样本进行学习以建立分类预测模型,再用测试集测试各个模型的分类准确度。结果:SVM分类器和DT分类器效果较好,准确率超过80%。结论:以反映血管弹性的脉搏波参数结合性别、年龄作为特征并使用SVM或者DT算法来构建分类预测模型,有一定实用价值。

This study aimed to compare the classifying ability of common classification algorithms in cerebral infarction prediction. Age, gender and 6 pulse wave indexes related with arterial elasticity were taken as the characteristics of each case. These cases were divided into two sets （the training set and the test set） at a ratio of 3 to 1. Several classifiers were constructed with the algorithms of ANN, Bayes, Decision Tree, k-NN and SVM. The modules of these classifiers were constructed by training the cases in the training set. The classification accuracy of these modules constructed with different algorithms was determined by testing each module with the cases in the test set. The classifiers constructed with SVM and DT had better performance （accuracy〉80%）. Taking the age, gender and 6 pulse wave indexes related with arterial elasticity as the case characteristics and using SVM or DT algorithm to construct modules for cerebral infarction prediction may have some practical potential.