Table 3.

Predicting the value of high-incidence days using machine learning in the training data of Tokyo.

	AUROC (95% CI)	Accuracy	Sensitivity	Specificity	F1-score
XGBoost	0.906 (0.868–0.944)	0.835 (0.806–0.864)	0.848 (0.778–0.918)	0.833 (0.796–0.871)	0.734 (0.682–0.786)
RF	0.904 (0.866–0.943)	0.838 (0.814–0.862)	0.838 (0.759–0.918)	0.841 (0.811–0.870)	0.735 (0.692–0.778)
LDA	0.903 (0.858–0.948)	0.842 (0.794–0.890)	0.826 (0.790–0.862)	0.849 (0.788–0.911)	0.740 (0.681–0.799)
LR	0.904 (0.856–0.952)	0.832 (0.788–0.875)	0.858 (0.778–0.937)	0.825 (0.762–0.888)	0.733 (0.675–0.791)
SVM	0.905 (0.858–0.952)	0.839 (0.797–0.880)	0.846 (0.793–0.899)	0.837 (0.786–0.888)	0.740 (0.691–0.789)
NB	0.901 (0.856–0.947)	0.843 (0.800–0.886)	0.831 (0.770–0.892)	0.848 (0.791–0.905)	0.742 (0.701–0.783)
MLP	0.904 (0.859–0.950)	0.834 (0.798–0.869)	0.849 (0.794–0.903)	0.829 (0.793–0.864)	0.734 (0.691–0.777)
KNN	0.899 (0.855–0.944)	0.839 (0.809–0.869)	0.832 (0.771–0.893)	0.840 (0.801–0.878)	0.736 (0.699–0.773)

Tokyo data (2005–2012) was analyzed.

AUROC area under the receiver operating characteristic curve, CI confidence interval, XGBoost extreme gradient boosting, RF random forest, LDA linear discriminant analysis, LR logistic regression, SVM support vector machine with radial basis function kernel, NB naïve Bayes, MLP multilayer perceptron, kNN k-nearest neighbors.