时间序列分析与机器学习方法在预测肺结核发病趋势中的应用

目的研究时间序列分析与机器学习方法在预测肺结核发病趋势中的应用。方法使用江苏省2009-2018年肺结核月度发病率数据,构建时间序列分析(ARIMA模型)、机器学习方法(支持向量回归(SVR)、BP神经网络)和两者的组合方法(ARIMA-SVR、ARIMA-BPANN)共5种预测模型,分析评价各模型预测性能。结果 ARIMA、SVR、BP神经网络、ARIMA-SVR、ARIMA-BPANN均方误差分别为0.0356、0.0364、0.0384、0.0329、0.0336;平均相对误差分别为5.76%、6.19%、6.20%、5.63%、5.70%。结论时间序列分析优于机器学习方法,而二者组合模型预测效果优于单独方法,ARIMA-SVR模型在江苏省肺结核发病趋势预测分析中具有较好的应用价值。

Application of Time Series Analysis and Machine Learning Methods in Predicting the Incidence of Tuberculosis

Objective To study the application of time series analysis and machine learning in predicting the incidence trend of tuberculosis.Methods Using the monthly incidence data of tuberculosis in Jiangsu Province from 2009 to 2018,five prediction models were constructed,including time series analysis(ARIMA),machine learning(support vector regression,BP neural network)and their combination methods(ARIMA-SVR,ARIMA-BPANN).The prediction performance of each model was analyzed and evaluated.Results The mean square errors of ARIMA,SVR,BP neural network,ARIMA-SVR and ARIMA-BPANN were 0.0356,0.0364,0.0384,0.0329 and 0.0336 respectively,and the average relative errors were 5.76%,6.19%,6.20%,5.63% and 5.70%,respectively.Conclusion Time series analysis is superior to machine learning method,and the combined model is superior to the single method.ARIMA-SVR model has a good application value in the prediction and analysis of tuberculosis incidence trend in Jiangsu Province.