Table 4.

Predictabilities of the prediction models for the number of heatstrokes of hospital admission and death cases among 6 models^a.

	GLM using WBGT only	GLM	GAM	RF	XGBoost	Consolidation of 16 GAMs specific to each cityb
	The number of heatstrokes of hospital admission and death cases
Overall predictive accuracies per city per 12 h
RMSE in training	0.68	0.62	0.62	0.3	0.44	0.61
RMSE in testing	1.14	0.92	0.83	1.09	1.08	1.42
Predictive accuracies on days when the number of heatstrokes spikedc
MAPE per 1-day (%)c in training	28.3	23.5	23.3	9.4	13.2	23.4
MAPE per 1-day (%)c in testing	37.7	23.7	10.6	21.2	24.9	10.4
Total absolute percentage error (%) in training	21.8	11.5	11.7	5.0	7.2	11.8
Total absolute percentage error (%) in testing	42.9	25.8	7.5	26.9	29.7	2.7

GLM generalized linear model, GAM generalized additive model, RF random forest, XGBoost extreme gradient boosting decision tree, WBGT wet bulb globe temperature, RMSE root-mean-square error, MAPE mean absolute percentage error.

^a Smaller RMSE, MAPE, and total absolute percentage error show better predictabilities.

^b Prediction models specific to each of the 16 cities were developed for city-specific prediction.

^c MAPE and total absolute percentage error were calculated after observed and predicted values were summed up per day (for MAPE) per the entire period (for total absolute percentage error) on days when the number of heatstrokes of hospital admission (i.e., moderate and severe cases) and death cases was 80th percentile (corresponding to 15.6 in 2015, 16 in 2016, 17 in 2017, and 23 in 2018) and over in each year. MAPE is a mean value of absolute errors divided by observed values.