关键词: 模型, 理论, 结核, 肺, 预测, 对比研究

Abstract:

Objective To compare the predictive effects of Autoregressive Integrated Moving Average (ARIMA) model, Elman neural network model and their combination (ARIMA-Elman) model on the incidence of pulmonary tuberculosis in Beijing, and explore the best predictive model.Methods The data between 2010 to 2017 was used as training data to establish three models (ARIMA, Elman, ARIMA-Elman). Data of 2018 was used as testing data to evaluate the predictive effects of three models. The Mean Absolute Error and the Mean Absolute Percentage Errors were two predictive efficiency indicators.Results The Relative Errors of ARIMA model, Elman neural network and ARIMA-Elman model were mostly within ±10% in predicting the monthly incidence of tuberculosis in Beijing, there were 8, 8 and 9, respectively. Except for the parts within ±10%, the number of Relative Errors within ±(10%-20%) for ARIMA model was 3 and 1 over ±20%; the number of Relative Errors within ±(10%-20%) for Elman neural network was 2 and 2 over ±20%; while the number of ARIMA-Elman combination model was 3 within ±(10%-20%). The mean Absolute Errors of the three models above were 44.7 (536/12), 47.8 (574/12) and 43.8 (526/12), while the Mean Absolute Percentage Errors were 8.7% (1.039/12×100%), 8.2% (0.99/12×100%) and 7.9% (0.953/12×100%) respectively. Both the two indicators of ARIMA-Elman model were smaller than the other two models.Conclusion ARIMA-Elman combination model has higher prediction accuracy and more desirable effect on the prediction of the incidence of tuberculosis in Beijing.

Key words: Models, theoretical, Tuberculosis, pulmonary, Forecasting, Comparative study