SARIMA与SVR模型在天津市肝炎发病率预测中的比较

Comparison of SARIMA and SVR Models in Predicting Hepatitis Incidence in Tianjin

摘要: 为比较乘积季节差分自回归移动平均模型(SARIMA)与支持向量机回归模型(SVR)对天津市病毒性肝炎发病率的预测效果，本文根据天津市2005年1月至2017年4月病毒性肝炎发病率数据建立SARIMA和SVR预测模型，对2017年5月至12月发病率预测。SVR模型预测的RMSE，MAE和MAPE分别为0.0767，0.0701和4.25%，SVR模型与SARIMA模型中最优模型相比，三个误差评价指标分别下降了0.1089，0.1008和6.04%。预测结果显示SVR模型预测效果优于SARIMA模型，将其用于天津市的病毒性肝炎发病率短期预测，有助于该地区病毒性肝炎的防治工作。

Abstract: To compare the seasonal autoregressive integrated moving average (SARIMA) model and support vector regression(SVR) model predicted effect on the incidence of viral hepatitis in Tianjin, we used data collected from January 2005 to April 2017 as training data while the data from May 2017 to December 2017 as testing data. The RMSE, MAE and MAPE predicted by the SVR model are 0.0767, 0.0701 and 4.25%, respectively. Compared with the SARIMA model of optimal model, the three error evaluation indexes of the SVR model decreased by 0.1089, 0.1008 and 6.04% severally. The prediction effect of SVR model is better than that of SARIMA model. Its application to the short-term prediction of the incidence of viral hepatitis in Tianjin is helpful to the prevention and treatment of viral hepatitis in this area.