**5. Construction of HFMD Early-Warning Model Based on Big Data**

The traditional HFMD epidemic warning model is mainly based on two major methods of time and space, namely the moving percentile prediction method and the spatial scanning statistical method. It also needs to use the repeated warning kicking algorithm to remove duplicates. The early-warning method is only based on historical data before and after the same period. It does not consider that the HFMD outbreak may be related to related weather and demographic factors. Blind warning and the warning process are cumbersome and wrong. There is a situation of blind warning every day and everywhere.

As there is no yet a complete and accurate early-warning mechanism for infectious diseases, HFMD epidemic early-warning methods based on adjustable parameters, moving percentiles and a combination of the two are proposed, mainly proposed different methods for the setting of HFMD early-warning threshold. The HFMD epidemic early-warning model is mainly based on the number of HFMD cases in the region generated by the above HFMD epidemic prediction model and the number of cases in the same period in history. That is, whether there will be an outbreak of HFMD epidemic in the early warning area, Or according to historical data in the same period, relevant health departments need to be warned to increase the vigilance of the HFMD epidemic in the region. The overall process of the model is shown in Figure 9.

*5.1. HFMD Epidemic Warning Model Based on Adjustable Parameters*

The parameters are adjustable, referring to different HFMD epidemic areas, according to the epidemic time of different seasons, the corresponding characteristic parameters in the HFMD prediction model can be dynamically adjusted according to the actual local conditions. The corresponding features can be obtained by the feature selection method, and the number of patients output by the feature parameters through the model is used as the early-warning threshold of HFMD infectious diseases. If the actual predicted value exceeds the early-warning threshold, the system will issue an early-warning signal. For example, when the minimum temperature, air humidity, the number of illnesses last week, the number of children aged 0–6, and the number of weeks are 24 degrees Celsius, 40%, 45 cases, 28 thousand people, and 16:00, through the HFMD epidemic prediction model, the incidence under this feature is obtained and used as the early-warning threshold. These characteristic values can be dynamically adjusted according to different regions and different times.

#### *5.2. HFMD Epidemic Warning Model Based on Historical Percentile Method*

First, establish a database of local historical cases of HFMD with the city as the unit, refer to the historical incidence of the same period in the past 3–5 years and two cycles before and after the same period. Like the forecast period, the general historical period is seven days. Then get the percentile (usually 80% after sorting from small to large) from the historical incidence as the early-warning threshold. When the predicted incidence in the statistical period is greater than this early-warning threshold, the system will automatically send an early-warning signal to relevant departments in the observation area within one day. For example, it is predicted that the weekly incidence of the area will reach 100. Among the nine incidence data of the same period and before and after the past three years, the incidence at the 80th percentile is 80. Then the system will send an early-warning signal to the relevant departments in the forecast area to remind the area that there may be an outbreak, or the need to strengthen prevention and control higher than the historical level.

#### *5.3. HFMD Epidemic Warning Model Based on Threshold Comparison*

Threshold comparison is to compare the two early-warning thresholds obtained by the above-mentioned parameter adjustable method and the historical percentile method, and use the smallest as the new early-warning threshold. The flow chart of the HFMD epidemic warning model based on threshold comparison is shown in Figure 10.

First, set the characteristic thresholds of the influencing factors of the HFMD epidemic, i.e., under the corresponding weather factors and demographic factors, the number of possible HFMD incidences is the number at risk of HFMD outbreaks, and these characteristic values are substituted into the HFMD prediction model to obtain the incidence number threshold.

Then, according to the local database of historical HFMD cases, calculate the 80th percentile number of cases in the same cycle and two swing cycles in the past three years, and use this value as another threshold for early-warning.This threshold is compared with the early-warning threshold obtained by parameter adjustment, and the minimum incidence threshold is used as the early-warning threshold of HFMD early-warning model.

Finally, the HFMD epidemic prediction model is used to predict the number of local cases. When the number of cases exceeds the early-warning threshold, the local health department will send an early-warning signal and take corresponding measures after verification. At the same time, it can feed back suggestions, continuously adjust the characteristic parameter thresholds, and improve the precise HFMD early-warning mechanism; When this incidence does not exceed the warning threshold, no warning signal is issued, only the number of HFMD infections that may occur in the local area.

**Figure 10.** Flow chart of HFMD epidemic warning model based on threshold comparison.
