Establishment of back-propagation neural network using meteorological data and incidence data of hand, foot and mouth disease

Objective In order to analysis the relationship between meteorological factors and the incidence of hand, foot and mouth disease (HFMD), we used weekly meteorological data and the weekly incidence data of HFMD to establish and evaluate the back-propagation neural network (BPNN) model.

Methods The incidence data of HFMD and meteorological data in Changning district of Shanghai from 2014 to 2017 were collected. Software SPSS 13.0 was used to analyze the relationship between the incidence HFMD and meteorological factors. The prediction model of BPNN was established with software Matlab 7.0. Mean error rate (MER) and determination coefficient (R²) were used to evaluate the fitting effect and the incidence data of HFMD in 2017 was used to verify the model.

Results The weekly incidence of HFMD was positively correlated with the average temperature, the highest air temperature, the lowest air temperature, cumulative precipitation, relative humidity, minimum relative humidity, average wind speed (all P＜0.05), but negatively correlated with the mean air pressure of the same week (r＝ –0.527, P＜0.001). The weekly incidence of HFMD was positively correlated with the average air temperature, the highest air temperature, the lowest air temperature, cumulative precipitation, relative humidity, minimum relative humidity, average wind speed (all P＜0.05), but negatively correlated to the mean air pressure of the past week (r＝ –0.522, P＜0.001). The fitting effect using the meteorological data of the past week (MER＝19.0%, R²＝0.895) was slightly better than that using the data of same week (MER＝20.2%, R²＝0.894). Verified by meteorological data of 2017, the absolute error between the predicted value and actual value was between 0 and 9 and the average was 2.52.

Conclusion The BPNN model established is suitable for the prediction of the incidence of HFMD in Changning.