基于百度指数的流感预测模型比较研究

Comparativestudy of influenza prediction models based on Baidu index

目的将流感关键字的百度指数与流感样病例(ILI)监测数据相结合,探讨建立最优流感预测模型。方法使用芜湖地区2014年第1周至2018年第52周数据建立模型,使用2019年第1周至2019年第10周数据评价模型。围绕流感的称谓、症状、治疗和预防确定流感关键字,分析不同滞后期流感关键字百度指数与流感样病例占门诊量百分比(ILI%)相关性,在相关程度最大的滞后期分别建立多重线性回归模型、主成分回归模型、回归和时间序列组合模型,比较各模型的拟合优度和预测误差。结果2014-2018年芜湖市国家流感监测哨点医院共上报ILI病例19180例,ILI%为2.75%,2014-2018年分别为1.57%、1.31%、2.48%、3.36%、4.98%。当百度指数与ILI%同步时相关程度最大,筛选出13个相关系数r≥0.3的关键字用于建立模型。评价结果显示,回归和时间序列的组合模型拟合优度和预测效果最好,R^2和调整R^2分别为0.840、0.829,平均相对误差为20.748%、14.893%。结论利用百度指数建立回归和时间序列组合模型可作为流感传统监测和预测的有益补充。

Objective To establish an optimal influenza prediction model by combining Baidu index of influenza keywords with the monitoring data of influenza-like cases(ILI).Methods Data from the first week of 2014 to the 52nd week of 2018 in Wuhu area were used to establish the model,and the data from the first week of 2019 to the 10th week of 2019 were used to evaluate the model.Determined keywords by revolving around the appellation,symptoms,treatment and prevention of influenza,analyzed correlation between Baidu index of influenza keywords in different time lags with percentage of outpatient service of influenza-like cases,respectively set up multiple linear regression models,principal component regression models and regression and time series models in the time lags the largest related degree,comparing the goodness of fit and predicted deviation of models.Results A total of 19180 cases of ILI were reported by Wuhu national influenza surveillance outpost hospital from 2014 to 2018,resulting ILI%of 2.75%,the annual ILI%were 1.57%,1.31%,2.48%,3.36%and 4.98%,respectively.The maximum related degree was observed by synchronization of Baidu Index and IIL%,13 keywords with correlation coefficient≥0.3 were selected to build the model.The evaluation showed the combination of regression and time series model had the best fit performance and predicted results,with R^2and adjusted R^2 of 0.840 and 0.829,and the average mean relative errors of 20.748%,14.893%・Conclusion Using Baidu index to establish regression and time series combined models can serve as a useful supplement to traditional monitoring and prediction of influenza.