Application of exponential smoothing method and ARIMA model to predicting pulmonary tuberculosis in students
ZHOU Mei-yan1, HUANG Ying2, YAN Yu-long1, ZHOU Jie1, HUANG Bo1
1. School of Public Health, University of South China, Hengyang, Hunan 421001, China; 2. Hengyang Municipal Center for Disease Control and Prevention, Hengyang, Hunan 421001, China
Abstract:Objective To explore the feasibility of exponential smoothing method and auto regressive integrated moving average (ARIMA) model in the prediction of pulmonary tuberculosis epidemic among students in Hengyang City, to compare the prediction effects of the two models and then determine a better model so as to provide references for early detection and scientific control of tuberculosis epidemic in schools. Methods We collected the data concerning pulmonary tuberculosis among students in Hengyang City from 2010 to 2020, and then compared the fit and advantages and disadvantages of prediction effects of the two models. Results The best fitting exponential smoothing model was the Holt-Winter additive model. The fitted R2, stationary R2, root mean square error (RMSE), mean absolute percentage error (MAPE), mean absolute error (MAE), and normalized Bayesian information criterion (BIC) were 0.666, 0.469, 5.716, 31.276, 3.873 and 3.606, respectively. Ljung-Box Q was 20.741, P 0.145, and the average relative error of prediction from January to December 2020 verified 39.98%. The best fitting ARIMA model was ARIMA(0,0,1)×(0,1,1)12 model, and the fitted stationary R2, R2, RMSE, MAPE, MAE and normalized BIC were 0.500, 0.603, 6.532, 34.623, 4.443 and 3.885, respectively. The average relative error of prediction from January to December 2020 verified was 120.76%. Conclusion Compared with the ARIMA model, the exponential smoothing model is more effective in fitting the number of pulmonary tuberculosis cases among students in Hengyang City; moreover, its prediction accuracy is higher.
[1] World Health Organization.Global tuberculosis report 2019[R].Geneva: WHO,2019:1-30. [2] 成君,夏倍惜,刘二勇,等.学校结核病突发疫情处置的思考[J].中国防痨杂志,2018,40(2):145-148. [3] 许俐颖,秦国友.ARIMA模型和指数平滑法在我国北方流感样病例预测中的应用和比较[J].实用预防学,2020,27(7):889-893. [4] 宋媛媛,王雷,熊甜,等.ARIMA模型与GM(1,1)模型在痢疾发病数预测中的比较研究[J].实用预防医学,2019,26(7):888-892. [5] 包娅薇,邵明, 陈雨婷,等.自回归求和滑动平均(ARIMA)模型在全球新型冠状病毒肺炎发病人数预测中的应用[J].中华疾病控制杂志,2020,24(5):543-548. [6] 王淑霞,李明阳,刘希波,等.ARIMA模型在全国艾滋病新发现病例数预测中的可行性研究[J].中国艾滋病性病,2020,26(7):705-708. [7] Caley M,Fowler T,Welch S. A risk of developing tuberculosis from a school contact: retrospective cohort study,United Kingdom,2009[J].Euro Surveillance, 2010,15(11):321-334. [8] Faccini M,Codecasa LR,Ciconali G,et al.Tuberculosis outbreak in a primary school,Milan,Italy[J].Emerg Infect Dis, 2013,19(3):485-487. [9] Centers for Disease Control and Prevention.Transmission of Mycobacterium tuberculosis in a high school and school-based supervision of an isoniazid-rifapentine regimen for preventing tuberculosis-Colorado,2011-2012[J].MMWR Morb Mortal Wkly Rep,2013,62(39):805-809. [10] 成君,刘剑君.我国学校结核病疫情监测和预警的现状与进展[J].中国防痨杂志,2020,42(5):436-441. [11] 白丽琼.是危机,还是机遇?——湖南省桃江县发生学校结核病聚集性疫情的思考[J].实用预防医学,2018,25(1):1-2. [12] 刘维华,朱建良,郭付爱,等.应用时间序列法建立学校结核病预警模型研究[J].中国学校卫生,2018,39(11):1682-1685. [13] 杨文姣,肖俊玲,丁国武.ARIMA模型和BP神经网络模型在甘肃省结核病发病率预测中的应用[J].中华疾病控制杂志,2019,23(6):728-732.