Comparison of single and hybrid models for predicting the incidence of hand, foot, and mouth disease in Changsha
10.3969/j.issn.1006-2483.2026.03.005
- VernacularTitle:单一模型与组合模型在长沙市手足口病发病预测中的应用比较
- Author:
Kun SUN
1
;
Shuilian CHEN
2
;
Jinsong QIU
2
;
Yinzhu ZHOU
2
Author Information
1. Changsha Center for Disease Control and Prevention (Changsha Health Comprehensive Supervision and Law Enforcement Bureau), Changsha , Hunan 410000, China;China Field Epidemiology Training Program, Chinese Center for Disease Control and Prevention (Chinese Academy of Preventive Medicine), Beijing 100050, China
2. Changsha Center for Disease Control and Prevention (Changsha Health Comprehensive Supervision and Law Enforcement Bureau), Changsha , Hunan 410000, China
- Publication Type:Journal Article
- Keywords:
Hand, foot and mouth disease;
Prediction;
SARIMAX;
BPNN;
LSTM;
Combination model
- From:
Journal of Public Health and Preventive Medicine
2026;37(3):24-28
- CountryChina
- Language:Chinese
-
Abstract:
Objective To compare the performance of seasonal difference autoregressive moving average exogenous variable model (SARIMAX), backpropagation neural network (BPNN), long short-term memory network (LSTM), and various combination models in fitting and predicting the incidence of hand, foot, and mouth disease (HFMD), and to provide a reference for optimizing HFMD prediction and prevention and control decisions. Methods Taking the monthly HFMD incidence data in Changsha from 2010 to 2024 as the training set, and the monthly HFMD incidence data from January to August 2025 as the validation set, the single models, SARIMAX-BPNN, SARIMAX-LSTM and weight combination models were constructed respectively. The R², mean absolute percentage error (MAPE), mean absolute error (MAE), and root mean square error (RMSE) were used to evaluate the model fitting and prediction performance. Results Based solely on MAPE, the LSTM model showed the best fitting and prediction performance. A comprehensive analysis of multiple indicators including R², MAPE, MAE, and RMSE indicated that the combination models had superior prediction performance. Compared with the LSTM, RMSE predicted by equal weight combination, different weight combination and SARIMAX-BPNN decreased by 40.97%, 37.50%, and 25.00%, respectively, while MAPE increased by 124.84%, 96.90%, and 89.69%, respectively. The fitting effect was as follows: LSTM > unequal weight combination > equal weight combination > SARIMAX-LSTM > SARIMAX-BPNN > BPNN > SARIMAX. The prediction performance was as follows: SARIMAX-BPNN > unequal weight combination > equal weight combination > LSTM > SARIMAX-LSTM > BPNN > SARIMAX. Conclusion SARIMAX-BPNN combination or weighted combination of multiple different models are two more robust combination strategies for predicting the incidence trend of hand, foot and mouth disease in Changsha.
