基于机器学习的郁证证型分类模型构建研究

Research on Construction of Classification Model of Depression Syndrome Based on Machine Learning

目的基于机器学习的随机森林和人工神经网络算法构建郁证证型分类模型,并采用混淆矩阵评估其模型的准确度。方法医案数据来自古今医案云平台、中国知网、万方、维普数据库,共纳入1010例医案,训练集和测试集划分比例为7∶3。利用Python在Jupyter notebook中进行特征提取,再通过随机森林、人工神经网络构建郁证证型分类模型,最后利用混淆矩阵验证分类结果准确性。结果利用随机森林算法构建的证型分类模型,整体准确度为89.44%,其中肝气郁结95.00%,气郁化火82.05%,痰气郁结89.29%,心神失养85.07%,心脾两虚89.74%,心肾阴虚95.16%;利用人工神经网络算法构建的证型分类模型,整体准确率为96.03%,其中肝气郁结100.00%,气郁化火92.31%,痰气郁结96.43%,心神失养91.04%,心脾两虚97.44%,心肾阴虚100.00%。结论两类分类模型分类结果均达到了较理想的效果,但人工神经网络分类模型准确度高于随机森林分类模型准确率,其非线性、模糊性等特征更适合于中医证型分类预测,可为今后中医诊断研究提供新的思路与方向。

Objective To construct the classification model of depression syndrome based on random forest and artificial neural network algorithm of machine learning,and use confusion matrix to evaluate the accuracy of the model.Methods The medical record data mainly came from ancient and modern medical record cloud platform,CNKI,Wanfang and VIP Databases,a total of 1010 medical records were included in this study.The proportion of training set and test set was 7∶3.The features was extracted by using Python in Jupyter notebook.Then the classification model of depression syndrome was constructed by random forest and artificial neural network,and finally the accuracy of the classification results was verified by confusion matrix.Results The overall accuracy of the syndrome classification model constructed by random forest algorithm was 89.44%,including liver Qi stagnation(95.00%),Qi stagnation fire(82.05%),phlegm Qi depression(89.29%),loss of mind(85.07%),heart and spleen deficiency(89.74%)and heart and kidney Yin deficiency(95.16%).The syndrome classification model constructed by artificial neural network algorithm had an overall accuracy of 96.03%,including liver Qi stagnation(100.00%),Qi stagnation fire(92.31%),phlegm Qi depression(96.43%),loss of mind(91.04%),heart and spleen deficiency(97.44%)and heart and kidney Yin deficiency(100.00%).Conclusion The classification results of the two classification models have achieved ideal results.However the accuracy of artificial neural network classification model is higher than that of random forest classification model,and its nonlinear and fuzzy characteristics are more suitable for the classification and prediction of traditional Chinese medicine(TCM)syndrome types.It can provide new ideas and directions for TCM diagnosis research in the future.