基于Prophet等时间序列季节模型的肺结核发病预测及对比分析

目的 探索河南省2014—2019年肺结核发病趋势及季节性特征，比较Prophet模型、ARIMA季节模型和Holt-Winters模型的拟合及预测效果，为肺结核防控提供科学依据。方法 基于河南省2014年1月至2018年12月肺结核月发病数据，建立Prophet模型、ARIMA季节模型和Holt-Winters模型，采用2019年1月至12月肺结核月报告发病数据验证预测效果。评价指标选取均方根误差（RMSE）、平均绝对百分比误差（MAPE）、平均绝对误差（MAE）、平均置信区间宽度、真实值超出置信区间个数。结果 模型拟合结果显示，河南省肺结核发病呈逐年下降趋势，每年3—5月达到发病高峰，2月和10月出现低谷；Prophet模型拟合及预测表现最优，评价指标RMSE、MAPE、MAE、平均置信区间宽度均低于另外两个模型，Holt-Winters模型次之，ARIMA(0,1,1)×(0,1,1)12模型拟合及预测效果相对较差。结论 Prophet模型具有较高的拟合预测准确度和精确度，可以很好地捕捉河南省肺结核发病趋势，模型拟合结果对肺结核防控工作具有一定指导意义。

Prediction and comparative analysis of tuberculosis incidence based on Prophet and other time series seasonal model

To explore the incidence trend and seasonal characteristics of tuberculosis in Henan province from 2014 to 2019, and to compare the fitting and prediction effects of Prophet model, seasonal ARIMA model and Holt-Winters model, so as to provide scientific basis for tuberculosis prevention and control. Methods Based on the monthly incidence data of tuberculosis in Henan province from January 2014 to December 2018, the Prophet model, Seasonal ARIMA model and Holt-Winters model were established by using R 4.0.3 and SPSS 26.0, and the prediction effect was verified by using the monthly incidence data of tuberculosis from January to December 2019. Root mean square error(RMSE), mean absolute percentage error(MAPE), mean absolute error(MAE), average confidence interval width and the number of real values exceeding the confidence interval were selected as evaluation indexes. Results The incidence of pulmonary tuberculosis in Henan province showed a downward trend year by year, and the peak appeared from March to May, and the trough appeared in February and October. The fitting and prediction performance of Prophet model was the best, the evaluation indexes RMSE, MAPE, MAE and Average confidence interval width were lower than the other two models, Holt-Winters model was the second best, ARIMA(0,1,1) ×(0,1,1) 12 was the worst. Conclusion The prediction model of Prophet has high accuracy and precision, which can well capture the trend of tuberculosis incidence in Henan province and predict the number of patients.