基于SARIMA-SVR模型的喀什地区流行性腮腺炎的流行趋势预测  

Prediction of epidemic trend of mumps in Kashgar area based on SARIMA-SVR model

作　　者：曾婷 谢娜[2] 热木孜亚·热布哈提 王凯[1] 王童敏 ZENG Ting;XIE Na;RAMZIYA Rifhat;WANG Kai;WANG Tong-min(School of Medical Engineering and Technology,Xinjiang Medical University,Urumqi,Xinjiang,830017,China)

机构地区：[1]新疆医科大学医学工程技术学院,新疆维吾尔自治区乌鲁木齐830017 [2]新疆维吾尔自治区疾病预防控制中心 [3]喀什地区疾病预防控制中心,新疆维吾尔自治区喀什市844000

出　　处：《现代预防医学》2021年第12期2139-2143,共5页Modern Preventive Medicine

基　　金：新疆自治区自然科学基金(2020D01A10)。

摘　　要：目的利用季节性自回归滑动平均模型(seasonal autoregressive integrated moving average,SARIMA)、支持向量回归模型(support vector regression,SVR)对喀什地区流行性腮腺炎(mumps)的月发病数进行预测,在上述两模型的基础上建立SARIMA-SVR组合模型,提高预测的精准度,为控制新疆喀什地区2021年流腮传播趋势提供科学预测。方法以喀什地区2005年1月—2017年12月的流腮月发病数据为训练集,进行数据的拟合以及预测模型的训练,分别建立SARIMA、SVR、SARIMA-SVR组合模型。对2018年1月—2020年12月的流腮月发病数进行预测,并与实际值相比较,采用均方根误差(root mean square error,RMSE)衡量模型预测性能。结果ARIMA(2,1,1)(0,0,1)12为最优的SARIMA模型,建立的SARIMA、SVR、SARIMA-SVR组合模型预测2018年1月—2020年12月的喀什地区流腮月发病数的RMSE分别为:9.611、9.545、3.427。结论SARIMA-SVR组合模型对喀什地区流腮的预测精度高于单一预测模型,故选取该模型建立方式,利用2005年1月—2020年12月的喀什地区流腮月发病数据预测该地区2021年的月发病数。Objective To predict the monthly incidence of epidemic mumps in Kashgar region using Seasonal Autoregressive Integrated Sliding Average(SARIMA) and Support Vector Regression(SVR) models. The combined SARIMA-SVR model was developed on the basis of the above two models to improve the accuracy of prediction and to provide scientific prediction for controlling the trend of mumps transmission in Kashgar, Xinjiang in 2021. Methods The monthly incidence data of mumps in Kashgar region from January 2005 to December 2017 were used as the training set, and the data were fitted and the prediction models were trained to establish SARIMA, SVR, and SARIMA-SVR combined models, respectively. The monthly incidence of mumps from January 2018 to December 2020 was predicted and compared with the actual values, and the root mean square error(RMSE) was used to evaluate the prediction performance of the models. Results ARIMA(2,1,1)(0,0,1)12 was the optimal SARIMA model, and the RMSEs of the established SARIMA, SVR, and combined SARIMA-SVR models for predicting the monthly incidence of mumps in Kashgar from January 2018 to December 2020 were 9.611, 9.545, and 3.427,respectively. Conclusion SARIMA-SVR combined model has higher prediction accuracy than single prediction model for mumps in Kashgar region, so this model is selected to build the model that use the monthly incidence data of mumps in Kashgar region from January 2005 to December 2020 to predict the monthly incidence of mumps in the region in 2021.

关 键 词：SARIMA模型 SVR模型 SARIMA-SVR组合模型 流行性腮腺炎 预测 

分 类 号：R512.1[医药卫生—内科学]