RNN、JPR及ARIMA 3种模型预测中国农村孕产妇死亡率的比较

尹小兰; 何芯芯; 杜林; 李远盛; 张俊辉

doi:10.16462/j.cnki.zhjbkz.2023.11.011

RNN、JPR及ARIMA 3种模型预测中国农村孕产妇死亡率的比较

doi: 10.16462/j.cnki.zhjbkz.2023.11.011

西南医科大学公共卫生学院流行病与卫生统计学教研室，泸州 646000

基金项目:

全国统计科学研究项目 2021LZ31

西南医科大学校级科研项目 2021ZKMS002

详细信息

通讯作者:
张俊辉，E-mail：vzjh960500@163.com

中图分类号: R181
计量
- 文章访问数: 460
- HTML全文浏览量: 99
- PDF下载量: 44
- 被引次数: 0
出版历程
- 收稿日期: 2022-07-13
- 修回日期: 2022-10-02
- 网络出版日期: 2023-11-20
- 刊出日期: 2023-11-10

A comparative study of the RNN, the JPR, and ARIMA for predicting maternal mortality ratio in rural areas in China

Department of Epidemiology and Health Statistics, School of Public Health, Southwest Medical University, Luzhou 646000, China

Funds:

National Statistical Science Research Project of China 2021LZ31

Scientific Research Project of Southwest Medical University 2021ZKMS002

More Information

Corresponding author: ZHANG Junhui, E-mail: vzjh960500@163.com

摘要

摘要: 目的利用循环神经网络(recurrent neural network, RNN)模型、Joinpoint回归(Joinpoint regression, JPR)模型、差分自回归移动平均(autoregressive integrated moving average, ARIMA)模型预测中国农村孕产妇死亡率(maternal mortality rate, MMR)的应用价值，并对《“健康中国2030”规划纲要》等文件提出的MMR下降目标能否实现进行统计预测。方法基于2000―2019年中国农村MMR数据建立RNN、JPR及ARIMA模型，通过平均绝对误差(mean absolute error，MAE)、均方误差(mean squared error，MSE)、均方根误差(root mean square error，RMSE)比较3种模型的回代拟合效果，通过残差及相对误差比较3种模型对2020年中国农村MMR数据进行点预测的效果，利用优选模型对2021―2030年中国农村的MMR进行预测。结果 2000―2020年中国农村的MMR整体呈持续下降趋势。3种模型回代拟合效果由好到差排序为：RNN>JPR>ARIMA，RNN模型的MAE、MSE、RMSE均＜0.02。3种模型的点预测效果由好到差排序为：RNN>JPR>ARIMA。最优的RNN模型预测结果显示，2022年中国农村MMR预测值为18.02/10万，能够实现《“健康中国2030”规划纲要》等文件的下降目标；2025及2030年中国农村MMR预测值分别为17.58/10万、17.27/10万，均不能实现《“健康中国2030”规划纲要》等文件的下降目标。结论 RNN模型预测效果优于JPR模型及ARIMA模型，JPR模型预测效果一般，ARIMA模型不太适合本数据的预测。
- 孕产妇死亡率 /
- 循环神经网络模型 /
- Joinpoint回归模型 /
- 差分自回归移动平均模型 /
- 预测
Abstract: Objective To explore the application value of the recurrent neural network (RNN) model, the joinpoint regression (JPR) model, and the autoregressive integrated moving average (ARIMA) model on predicting maternal mortality ratio (MMR) in rural areas in China, and to make statistical predictions on whether the MMR decrease targets of "Healthy China 2030" and other documents will be achieved or not. Methods The RNN, JPR, and ARIMA models were constructed based on the data of MMR in rural areas in China from 2000 to 2019, and the mean absolute error (MAE), mean squared error (MSE) and root mean squared error (RMSE) were used to compare the back generation fitting effects of the three models. The residuals and relative errors were used to compare the point prediction effects of the three models for MMR data in rural areas in 2020. Finally, the optimal model was selected to forecast the MMRs in rural areas from 2021 to 2030. Results From 2000 to 2020, the MMRs in rural areas in China showed an overall continuous downward trend. The back generation fitting effects of the three models were ranked in descending order as follows: RNN > JPR > ARIMA, and the MAE, MSE, and RMSE of the RNN models were all less than 0.02. The accuracies of the three models for the point prediction of rural MMR in 2020 were ranked in descending order as follows: RNN > JPR > ARIMA. The prediction results of the optimal RNN model showed that the rural MMR in 2022 would be 18.02/100 000, indicating the decreased target of the relevant documents would be achieved in 2022. The MMR in 2025 and 2030 would be 17.58/100 000 and 17.27/100 000, respectively, indicating the decrease targets of the "Health China 2030" and other documents would not be achieved in 2025 and 2030. Conclusions The predictive performance of the RNN model is much better than those of the JPR model and the ARIMA model. The JPR model is an acceptable predictive model, while the ARIMA model is less suitable for the prediction of this data.
- Maternal mortality ratio /
- Recurrent neural network model /
- Joinpoint regression model /
- Autoregressive integrated moving average /
- Prediction

HTML全文

图 1 2000―2020年中国农村孕产妇死亡率的变化趋势

Figure 1. Trends of maternal mortality ratios in rural areas in China from 2000 to 2020

下载: 全尺寸图片幻灯片

图 2 3种模型的回代拟合效果

RNN: 循环神经网络；ARIMA: 差分自回归移动平均。

Figure 2. Effects of back generation fitting of the three models

RNN: recurrent neural network; ARIMA: autoregressive integrated moving average.

下载: 全尺寸图片幻灯片

表 1 3种模型的回代拟合及点预测效果比较

Table 1. Comparison the effects of back generation fitting and point prediction of the three models

模型Model	回代拟合Back generation fitting			点预测Point prediction
模型Model	MAE	MSE	RMSE	残差 Residuals	相对误差/% Relative errors/%
循环神经网络模型Recurrent neural network model	0.011 3	0.000 2	0.013 6	0.04	0.22
Joinpoint回归模型Joinpoint regression model	0.875 7	0.983 2	0.991 6	-0.90	4.86
ARIMA模型ARIMA model	2.337 4	10.068 6	3.173 1	-1.88	10.10
注：ARIMA, 差分自回归移动平均; MAE, 平均绝对误差; MSE, 均方误差; RMSE, 均方根误差。 Note: ARIMA, autoregressive integrated moving average; MAE, mean absolute error; MSE, mean squared error; RMSE, root mean squared error.

下载: 导出CSV

参考文献(14)

[1]	张天成, 陈露, 谭利明, 等. 中国孕产妇死亡率时空变化及预测探究[J]. 中国卫生统计, 2018, 35(5): 745-747. DOI: 10.3969/j.issn.1002-3674.2018.05.028. Zhang TC, Chen L, Tan LM, et al. Exploring the spatial and temporal variation and prediction of maternal mortality in China[J]. Chinese Journal of Health Statistics, 2018, 35(5): 745-747. DOI: 10.3969/j.issn.1002-3674.2018.05.028.
[2]	周红英, 邓峰, 吕菊红. 1991~2016年中国孕产妇死亡率变化情况分析[J]. 中国计划生育和妇产科, 2019, 11(6): 34-36, 44. DOI: 10.3969/j.issn.1674-4020.2019.06.10. Zhou HY, Deng F, Lyu JH. Analysis on the changes of maternal mortality in China from 1991 to 2016[J]. Chinese Journal of Family Planning & Gynecotokology, 2019, 11(6): 34-36, 44. DOI: 10.3969/j.issn.1674-4020.2019.06.10.
[3]	刘兴会, 陈锰. 降低中国可避免的孕产妇死亡[J]. 中国实用妇科与产科杂志, 2020, 36(1): 54-56. DOI: 10.19538/j.fk2020010113. Liu XH, Chen M. Reducing avoidable maternal deaths in China[J]. Chin J Pract Gynecol and Obstet, 2020, 36(1): 54-56. DOI: 10.19538/j.fk2020010113.
[4]	陈露. 中国孕产妇死亡率的时间和空间变化趋势及其模型预测研究[D]. 衡阳: 南华大学, 2019. Chen L. Time and spatial variation trends and model prediction of maternal mortality in China[D]. Hengyang: South China University, 2019.
[5]	Zhang JB, Zheng Y, Qi DK. Deep spatio-temporal residual networks for citywide crowd flows prediction[C]. New York: ACM, 2017.
[6]	杨之洵. 中国胃癌负担趋势及预测研究[D]. 北京: 北京协和医学院, 2019. Yang ZX. Trend and prediction of burden of stomach cancer in China[D]. Beijing: Peking Union Medical College, 2019.
[7]	刘洁, 高茵茵, 曲波, 等. 应用ARIMA模型预测中国孕产妇死亡率[J]. 中国医科大学学报, 2011, 40(2): 107-108, 121. DOI: 10.1631/jzus.B1000265. Liu J, Gao YY, Qu B, et al. Study of the feasibility for application of ARIMA model to predict maternal mortality ratio in China[J]. J Chin Med Univ, 2011, 40(2): 107-108, 121. DOI: 10.1631/jzus.B1000265.
[8]	庞艳蕾. 孕产妇、婴儿及5岁以下儿童死亡率分析与预测[D]. 潍坊: 潍坊医学院, 2016. Pang YL. Study on analysis and prediction of maternal mortality rate, infant mortality rate, child mortality rate under age 5[D]. Weifang: Weifang Medical College, 2016.
[9]	马倩倩, 何贤英, 崔芳芳, 等. 基于ARIMA与NNAR模型的中国食管癌疾病负担预测[J]. 中华疾病控制杂志, 2021, 25(9): 1048-1053. DOI: 10.16462/j.cnki.zhjbkz.2021.09.010. Ma QQ, He XY, Cui FF, et al. Prediction of disease burden of esophageal cancer in China based on ARIMA and NNAR models[J]. Chin J Dis Control Prev, 2021, 25(9): 1048-1053. DOI: 10.16462/j.cnki.zhjbkz.2021.09.010.
[10]	王雅文, 沈忠周, 严宝湖, 等. ARIMA模型和ARIMA-GRNN模型在AIDS发病预测中的应用[J]. 中华疾病控制杂志, 2018, 22(12): 1287-1290. DOI: 10.16462/j.cnki.zhjbkz.2018.12.020. Wang YW, Shen ZZ, Yan BH, et al. Application of ARIMA and hybrid ARIMA-GRNN models in forecasting AIDS incidence in China[J]. Chin J Dis Control Prev, 2018, 22(12): 1287-1290. DOI: 10.16462/j.cnki.zhjbkz.2018.12.020.
[11]	郑英. 孕产妇死亡率、儿童死亡率水平及影响因素分析[D]. 大连: 大连医科大学, 2007. Zheng Y. Level and effect of maternal mortality rate and child mortality rate[D]. Dalian: Dalian Medical University, 2007.
[12]	郑超. 下一站健康——中国妇女发展基金会"母亲健康快车"项目综述[J]. 中国社会组织, 2014(3): 35-37. DOI: 10.3969/j.issn.2095-4786.2014.03.014. Zheng C. The next stop for health: An overview of the China Women's Development Foundation's "Mother's Health Express" project[J]. Chin Soc Organ, 2014(3): 35-37. DOI: 10.3969/j.issn.2095-4786.2014.03.014.
[13]	俞跃萍, 赵钟鸣, 刘锦桃. 世界和中国孕产妇死亡变化趋势及终身风险[J]. 中国公共卫生, 2019, 35(1): 53-57. DOI: 10.11847/zgggws1118295. Yu YP, Zhao ZM, Liu JT. Trends, levels and lifetime risks of maternal mortality in the world and in China[J]. Chin J Public Health, 2019, 35(1): 53-57. DOI: 10.11847/zgggws1118295.
[14]	刘博, 王明烁, 李永, 等. 深度学习在时空序列预测中的应用综述[J]. 北京工业大学学报, 2021, 47(8): 925-941. DOI: 10.11936/bjutxb2020120037. Liu B, Wang MS, Li Y, et al. Deep learning for spatio-temporal sequence forecasting: a survey[J]. J Beijing Univ Technol, 2021, 47(8): 925-941. DOI: 10.11936/bjutxb2020120037.