摘要
目的比较几种传统模型及机器学习方法,在甘肃省预测梅毒发病率的效果,并对未来发病率进行预测,为制定控制措施提供依据。方法应用MATLAB 2014a软件,对甘肃省2004-2015年梅毒发病率数据分别建立多项式回归、平滑样条插值、灰色系统GM(1,1)、自回归整合移动平均(ARIMA)、人工神经网络(ANN)和支持向量机(SVR)等数学模型,然后根据2016年实际发病率数据来检验预测效果以选择最佳预测模型,最后使用该模型预测2017-2020年发病率。结果构建的一次多项式、二次多项式、平滑样条方法、GM(1,1)、ARIMA、ANN和SVR模型,拟合2004-2015年梅毒发病率平均相对误差分别为20.04%、22.44%、8.10%、24.89%、11.00%、17.61%和24.72%,以平滑样条最小。7种模型预测2016年梅毒发病率,以ARIMA模型最佳,使用该模型预测2017-2020年发病率分别为19.11/10万、18.21/10万、18.57/10万和19.94/10万。结论不同数学模型拟合和预测效果不同,应根据实际数据选择合适的模型;ARIMA模型预测甘肃省近年梅毒发病率性能较好,预测2017-2020年发病率较为稳定。
Objective To compare the performance of traditional methods and machine learning methods in forecasting the incidence trend of syphilis,in order to facilitate the control of syphilis in Gansu province.Methods Polynomial regression,smoothing spline,GM（1,1）model,ARIMA model,ANN model and SVR model were used to analyze the syphilis data from 2004 to 2015 by MATLAB2014a,to study the incidence trend of syphilis.The predictive results were verified by the actual syphilis incidence in 2016,to find the most suitable model,which would be chosen to predict the incidence trend of syphilis in Gansu during 2017-2020.Results The mean absolute percentage error（MAPE）of first-and second-polynomila,smoothing spline,GM（1,1）,ARIMA,ANN and SVR models were 20.04%,22.44%,8.10%,4.89%,11.00%,17.61% and 24.72% respectively.The prediction of ARIMA was the closest to the real value of syphilis in 2016,showing 19.11,18.21,18.57 and 19.94 per hundred thousand for 2017-2020.Conclusion The ARIMA model has the best performance,and morbidity rate of syphilis will change steadily in the following years.
出处
《中国艾滋病性病》
CAS
CSCD
北大核心
2017年第7期647-650,共4页
Chinese Journal of Aids & STD
基金
甘肃省卫生行业科研计划资助项目(GSWSKY-2014-22)
关键词
梅毒
发病率
数学模型
预测
Syphilis
Incidence rate
Mathematical models
Prediction
