基于ARIMA模型和ARIMA-SVM组合模型的流行性感冒的发病预测研究

Influenza disease prediction based on ARIMA and ARIMA-SVM combination models

目的探讨ARIMA-SVM组合模型在流感发病预测中的应用,并与单纯ARIMA模型的预测效果比较。方法利用2017—2022年北京市流感发病数据拟合建立ARIMA模型和ARIMA-SVM组合模型,对2023年流感发病进行预测,并与实际流感数据进行验证比较,评价模型的预测效果。结果北京市2017年1月—2023年12月共报告流感病例报告数1250797例,月均发病14890例。构建最佳的ARIMA模型的为ARIMA(6,0,6)(0,1,2)365,模型预测相对误差范围在0.01%~165.62%之间,RMSE=570.07,MAPE=157.36%。ARIMA-SVM模型预测相对误差在0.00%~18.87%之间,RMSE=0.26,MAPE=1.90%。组合模型预测结果较单一ARIMA模型精度高。结论ARIMA与SVM联合模型对流感发病的拟合精度优于单一ARIMA模型,可用于流感发病的短期预测,组合模型不仅考虑了传染病发病数据的周期性特点,又克服了小样本、非线性的缺点,亦可推广到其他的传染病的发病预测,为传染病的预测、疾病控制以及资源的配置利用提供政策支持。

Objective To explore the application of the autoregressive integrated moving average(ARIMA)model and ARIMA-SVM combination model in predicting influenza incidence,and to compare their predictive performance.Methods Using influenza incidence data from 2011 to 2023 in Beijing,we established both ARIMA and ARIMA-SVM combination models to predict the incidence of influenza in 2023.The predictive performance of the models was evaluated by comparing the predicted values with the actual influenza incidence.Results From January 2017 to December 2023,a total of 1250797 influenza cases were reported in Beijing,with an average of 14890 cases per month.The best-fitting ARIMA model was ARIMA(6,0,6)(0,1,2)365.The relative error of this model’s predictions ranged from 0.01%to 165.62%,with an RMSE of 570.07 and a MAPE of 157.36%.The ARIMA-SVM combination model had a relative error ranging from 0.00%to 18.87%,with an RMSE of 0.26 and a MAPE of 1.90%.The combination model demonstrated higher accuracy than the ARIMA model alone.Conclusions The ARIMA-SVM combination model outperforms the ARIMA model alone in fitting influenza incidence data.It can be used for shortterm prediction of influenza incidence.The combination model not only accounts for the periodic characteristics of infectious disease data,but also overcomes the limitations of small sample sizes and non-linearity.It can be extended to predict the incidence of other infectious diseases,providing valuable support for disease prediction,control,and resource allocation.