Seasonal and Meteorological Drivers of Hand, Foot, and Mouth Disease Outbreaks Using Data-Driven Machine Learning Models

Abstract

Hand, Foot, and Mouth Disease (HFMD) predominantly affects children under the age of five and remains a significant public health concern in the Asia-Pacific region. HFMD outbreaks are closely linked to seasonal changes and meteorological factors, particularly in tropical and subtropical areas. In Thailand, a total of 657,570 HFMD cases were reported between 2011 and 2022 (12 years). This study aimed to identify the high- and low-risk HFMD outbreak areas using machine learning models: Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), Random Forests (RF), Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (XGBoost). Our findings showed that the XGBoost model outperformed the other models in predicting unseen data and defining the best model. The best model can be used to detect high-risk outbreak areas and to explore the relationship between meteorological factors and HFMD outbreaks. The results highlighted the seasonal distribution of high-risk HFMD outbreak months across different provinces in Thailand, with average maximum temperature, average rainfall, and average vapor pressure identified as the most influential factors. Furthermore, the best model was used to analyze HFMD outbreaks during the COVID-19 pandemic, showing a notable reduction in high-risk outbreak months and areas, likely due to the control measures implemented during this period. Overall, our model shows great potential as a tool for warnings, providing useful insights to help public health officials reduce the impact of HFMD outbreaks.

Keywords: and mouth disease; foot; hand; machine learning; meteorological factors; outbreak detection.

Figure 1

The ML framework for HFMD outbreak detection. Six models used Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), Random Forests (RF), Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (XGBoost).

Figure 2

Seasonal (A) line plot and (B) Polar plot for monthly HFMD incidence rates averaged across all provinces in Thailand during 2011–2022.

Figure 3

Maps of the proportion of high-risk HFMD outbreaks each month across 12 years (2011–2022), showing the seasonality and the distribution of the HFMD epidemics. The intensity of the pink color represented the proportion of high-risk HFMD outbreaks.

Figure 4

ROC curves predicting HFMD outbreaks for the training set (2011–2018).

Figure 5

ROC curves predicting high-risk HFMD outbreaks for the test set (2019–2022) from (A) RF and (B) XGBoost models.

Figure 6

Variable importance rankings for (A) RF and (B) XGBoost models.

Figure 7

The error of predicted HFMD outbreaks across 77 provinces (A) and the heatmap of predicted HFMD outbreak months (B). Positive errors represent overprediction (predicted high-risk outbreaks higher than actual), while negative errors represent underprediction (predicted high-risk outbreaks lower than actual). The grey highlighted area shows the duration of the lockdown control measure [44,45].