Table 2.

Classification Performance of Machine Learning Models to Predict Laser Corneal Refractive Surgery Option Via 10-Fold Cross-Validation

	Accuracy (%) (95% CI)	RCI (95% CI)	κ (95% CI)	P Valuea
Trained with SMOTE
Multiclass XGBoost	82.1 (81.1–83.0)	0.537 (0.525–0.549)	0.758 (0.747–0.769)	Reference
One-versus-rest XGBoost	81.7 (80.7–82.6)	0.531 (0.519–0.543)	0.753 (0.742–0.764)	0.578
One-versus-one XGBoost	81.9 (80.9–82.8)	0.534 (0.522–0.546)	0.756 (0.745–0.767)	0.780
Random forest	81.5 (80.5–82.4)	0.527 (0.515–0.539)	0.750 (0.739–0.761)	0.407
One-versus-rest SVM	75.3 (74.2–76.3)	0.422 (0.410–0.434)	0.668 (0.656–0.680)	<0.001
One-versus-one SVM	75.7 (74.7–76.7)	0.428 (0.415–0.441)	0.674 (0.662–0.686)	<0.001
DAG SVM	75.5 (74.5–76.5)	0.425 (0.412–0.438)	0.671 (0.659–0.683)	<0.001
Artificial neural network	76.0 (74.9–77.0)	0.432 (0.419–0.445)	0.677 (0.665–0.689)	<0.001
Trained without SMOTE
Multiclass XGBoost	80.2 (79.2–81.2)	0.514 (0.502–0.526)	0.730 (0.719–0.741)	0.011
One-versus-rest XGBoost	80.1 (79.1–81.1)	0.513 (0.501–0.525)	0.727 (0.715–0.738)	0.015
One-versus-one XGBoost	78.5 (77.4–79.5)	0.505 (0.493–0.517)	0.721 (0.709–0.732)	0.001
Without anticipated surgery option
Multiclass XGBoost	70.3 (69.2–71.4)	0.407 (0.394–0.420)	0.593 (0.581–0.605)	<0.001
One-versus-rest XGBoost	68.8 (67.7–69.9)	0.385 (0.372–0.398)	0.571 (0.559–0.583)	<0.001
One-versus-one XGBoost	68.3 (67.2–69.4)	0.380 (0.366–0.393)	0.565 (0.552–0.568)	<0.001

CI, confidence interval; RCI, relative classifier information; SVM, support vector machine.

^aComparison of accuracy with the best machine learning technique (multiclass XGBoost with SMOTE).