柳州市手足口病发病率预测模型的建立与评估

发布时间：2018-07-29 16:17

【摘要】：研究目的分析2010-2016年柳州市手足口病月发病率的时间分布规律,探讨季节性ARIMA模型、ARIMA-GRNN组合模型及XGBoost机器学习模型在手足口病发病率短期预测方面的应用。通过MAE、RMSE和MAPE等指标评估模型的优劣,从中选择出最适合柳州市手足口病月发病率预测的模型。探索性地将XGBoost机器学习方法应用于手足口病的短期预测,为传染病发病率的精准预测提供新的思路,为制定柳州市手足口病预防预警策略提供科学依据。研究方法使用Excel整理2010-2016年柳州市手足口病数据,用流行病学的方法描述其时间分布特征。将2010-2015年手足口病的月发病率数据作为训练样本,将2016年手足口病的月发病率数据作为预测样本,建立季节性ARIMA模型、ARIMA-GRNN组合模型和XGBoost机器学习预测模型。同时评估模型的优劣,评价指标为MAE、RMSE和MAPE,其值越小说明模型预测效果越好。使用IBM SPSS Statistics 16.0建立季节性ARIMA时间序列模型;使用Matlab R2014a软件在ARIMA模型的基础上建立ARIMA-GRNN组合预测模型;使用R version 3.3.3软件建立XGBoost机器学习模型。研究结果(1)2010-2016年柳州市各年手足口病累积报告发病率依次为494.72/10万、475.26/10万、618.90/10万、675.41/10万、852.08/10万、780.52/10万、765.44/10万。其中,7年累积手足口病重症病例729例,死亡40例。主要发病高峰为每年4-7月份,5月份最为严重,发病次高峰为每年的9-10月份。(2)柳州市手足口病月发病率季节性ARIMA预测模型表达式为ARIMA(1,0,0)x(1,1,0)12。模型参数AR(1)和SAR(1)均有统计学意义(P0.001),标准化BIC值为6.236,残差序列为白噪声(Q=8.359,P=0.937)。训练样本的MAE、RMSE、MAPE依次为13.581、20.407、38.322%;预测样本的MAE、RMSE、MAPE依次为11.962、9.481、32.689%。(3)柳州市手足口病月发病率ARIMA-GRNN组合预测模型的最佳光滑因子为0.030,所对应的的RMSE值为4.285。训练样本的MAE、RMSE、MAPE依次为10.393、13.443、36.160%;预测样本的MAE、RMSE、MAPE依次为9.481、11.863、30.101%。(4)柳州市手足口病月发病率XGBoost机器学习模型的最重要预测因素为lag24,其所对应的增益值、Cover值、频率值依次为0.505、0.187、0.093。训练样本的MAE、RMSE、MAPE依次为2.106、5.311、2.650%;预测样本的MAE、RMSE、MAPE依次为7.134、10.224、13.253%。研究结论(1)2010-2016年柳州市手足口病疫情较为严峻。发病情况有明显的季节性特征,呈现出一种双峰的趋势。(2)ARIMA(1,0,0)x(1,1,0)12模型、ARIMA-GRNN组合预测模型以及XGBoost机器学习模型均可用于柳州市手足口病月发病率的拟合和预测。(3)本研究建立的柳州市手足口病月发病率预测模型,ARIMA-GRNN组合预测模型的预测效果优于乘积季节性ARIMA模型,而XGBoost机器学习模型的预测效果则优于ARIMA模型和ARIMA-GRNN组合预测模型。
[Abstract]:Objective to analyze the temporal distribution of monthly morbidity of HFMD in Liuzhou from 2010 to 2016 and to explore the application of seasonal ARIMA model ARIMA-GRNN combined model and XGBoost machine learning model in short-term prediction of HFMD incidence. According to the evaluation model of mae RMSE and MAPE, the most suitable model for predicting the incidence of HFMD in Liuzhou was selected. The method of XGBoost machine learning is applied to the short-term prediction of HFMD, which provides a new idea for accurate prediction of incidence of infectious diseases and provides scientific basis for formulating the prevention and warning strategy of HFMD in Liuzhou. Methods the data of hand-foot-mouth disease from 2010 to 2016 in Liuzhou were collected by Excel and the time distribution characteristics were described by epidemiological method. The monthly morbidity data of hand, foot and mouth disease from 2010 to 2015 were used as training samples, and the monthly incidence data of hand foot and mouth disease in 2016 were used as prediction samples. The seasonal ARIMA model ARIMA-GRNN combined model and XGBoost machine learning prediction model were established. At the same time, the advantages and disadvantages of the model are evaluated. The smaller the value of MAE RMSE and MAPE is, the better the prediction effect of the model is. The seasonal ARIMA time series model is established by using IBM SPSS Statistics 16.0, the ARIMA-GRNN combination prediction model is built on the basis of ARIMA model by Matlab R2014a software, and the XGBoost machine learning model is established by using R version 3.3.3 software. Results (1) the cumulative reported incidence of HFMD in Liuzhou City from 2010 to 2016 was 494.72 / 100 000, 475.26 / 100 000, 618.90 / 100 000, 675.41 / 100 000 / 100000, 852.08 / 100 000 or 7,80.522 / 100 000 or 765.44 / 100, 000 respectively. Among them, 729 cases and 40 cases died of hand, foot and mouth disease in 7 years. The main peak was from April to July, and the second peak was from September to October. (2) the seasonal ARIMA prediction expression of monthly incidence of HFMD in Liuzhou was ARIMA (1 0 0) x (1 1 0) 12. The model parameters AR (1) and SAR (1) were statistically significant (P0. 001), the standardized BIC value was 6. 236, and the residual sequence was white noise. The Mae RMSE MAPE of the training samples was 13.581n 20.4070.38.322, and that of the predicted samples was 11.9629.481k32.689. (3) the best smoothing factor of the ARIMA-GRNN combination forecasting model for HFMD incidence in Liuzhou was 0.030, and the corresponding RMSE value was 4.285. The Mae RMSE MAPE of the training samples was 10.393 / 13.443 / 36.160, and that of the predicted samples was 9.481C / 11.863 / 30.101respectively. (4) the most important predictor of the XGBoost machine learning model for the monthly incidence of HFMD in Liuzhou was lag24, and the corresponding gain value / cover value and the frequency value were 0.5050.1870.093 and 0.5050.1870.093 respectively. The Mae RMSE MAPE of the training samples was 2.106, 5.3112.650, and the predicted mae RMSE MAPE of the training samples was 7.134 and 10.2240.13.253 respectively. Conclusion (1) the epidemic situation of hand, foot and mouth disease in Liuzhou from 2010 to 2016 is severe. The incidence has obvious seasonal characteristics. There is a bimodal trend. (2) ARIMA (1 0 0) x (1 0) 12 model / ARIMA-GRNN combined prediction model and XGBoost machine learning model can be used to fit and predict the monthly incidence of HFMD in Liuzhou. (3) the monthly incidence of HFMD in Liuzhou is predicted by this study. The prediction effect of ARIMA-GRNN combined forecasting model is better than that of product seasonal ARIMA model. The prediction effect of XGBoost machine learning model is better than that of ARIMA model and ARIMA-GRNN combination model.
【学位授予单位】：广西医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R181.3;R512.5

【相似文献】