基于随机森林的帕金森疾病诊断模型构建研究

来源：计算机技术与发展杂志2023年第04期北京时间：

作者：王小科;晏峻峰*

单位：湖南中医药大学 信息科学与工程学院,湖南 长沙 410208 Author(s): WANG Xiao-ke;YAN Jun-feng School of Information Science and Engineering,Hunan University of Chinese Medicine,Changsha 410208,China 关键词: SVM SMOTE;信息增益;随机森林;网格搜索;交叉验证 Keywords: SVM SMOTE;information gain;random forest;grid search;cross-validation 分类号: TP301 DOI: 10. 3969 / j. issn. 1673-629X. 2023. 04. 023 摘要: 近些年来,根据帕金森疾病( PD) 患者的语音数据对该疾病做出诊断成为一种行之有效的疾病诊断方法。 首先,针对语音数据集中存在非均衡数据和噪声样本的问题使用 SVM SMOTE 过采样技术,利用支持向量机分类器寻找支持向量并在此基础上合成新的样本以达到均衡数据集的目的;为了减少数据维度,降低学习难度,运用信息增益特征选择对所有特征属性计算数值并划分数据集以此来获得信息增益,根据信息增益的大小排序选取得到八个特征作为最优特征组合;最后,构建随机森林帕金森疾病诊断模型,并采用网格搜索和交叉验证相结合的方式进行参数调优,进一步优化模型,实现诊断模型准确率的进一步提高。 实验结果表明,优化后模型的准确率、灵敏度和特异度均有提升,为别为 96. 59% 、94. 81% 和 95. 49% ,且准确率均高于支持向量机、最邻近节点算法、朴素贝叶斯和决策树等决策模型。 Abstract: In recent years,the diagnosis of Parkinson’s disease based on the speech data of patients with Parkinson’s disease has becomean effective disease diagnosis method. Firstly,we use the SVM SMOTE oversampling technique to solve the problem of unbalanced dataand noise samples in the speech data set,and use the?SVM classifier to find the support vector and synthesize new samples on this basis toachieve the purpose?of the balanced data set. In order to reduce the data dimension and the difficulty of learning,we use the informationgain feature selection to calculate the value of all feature attributes and divide the data set to?obtain the information gain,and select eightfeatures as the optimal feature combination according to the?size of the information gain. Finally,a random forest Parkinson’s disease diagnosis model is constructed,and the parameters are optimized by the combination of grid search and cross-validation to further optimizethe model,so as to further improve the accuracy of the diagnosis model. The experimental results show that?the accuracy,sensitivity andspecificity of the optimized model are improved,which are 96. 59% ,94. 81% and?95. 49% ,respectively,and the accuracy is higher thanthat of support vector machine,nearest neighbor algorithm,naive Bayes and decision trees.

摘要：近些年来,根据帕金森疾病( PD) 患者的语音数据对该疾病做出诊断成为一种行之有效的疾病诊断方法。 首先,针对语音数据集中存在非均衡数据和噪声样本的问题使用 SVM SMOTE 过采样技术,利用支持向量机分类器寻找支持向量并在此基础上合成新的样本以达到均衡数据集的目的;为了减少数据维度,降低学习难度,运用信息增益特征选择对所有特征属性计算数值并划分数据集以此来获得信息增益,根据信息增益的大小排序选取得到八个特征作为最优特征组合;最后,构建随机森林帕金森疾病诊断模型,并采用网格搜索和交叉验证相结合的方式进行参数调优,进一步优化模型,实现诊断模型准确率的进一步提高。 实验结果表明,优化后模型的准确率、灵敏度和特异度均有提升,为别为 96. 59% 、94. 81% 和 95. 49% ,且准确率均高于支持向量机、最邻近节点算法、朴素贝叶斯和决策树等决策模型。

关键词：SVM SMOTE;信息增益;随机森林;网格搜索;交叉验证