Design of a Clinical Decision Support System for Fracture Prediction Using Imbalanced Dataset

Table 2

Comparisons of predictive performance between models designed based on the methods proposed in this study and other studies [40] with the CoIL challenge dataset [54].

CDSS model	AC (%)	SE (%)	SP (%)	G-mean	AUC
SVM + 100% oversampling	50.08	66.39	49.04	0.5706	0.5772
MLP + SMOTE	82.48	34.87	85.49	0.5460	0.6018
Hybrid SVM-MLP + 100% oversampling	52.1	63.87	51.36	0.5727	0.5762
LR + SMOTE	72.4	56.3	73.42	0.6429	0.6486
Hybrid SVM-LR + 100% oversampling	50.18	66.39	49.15	0.5712	0.5777
Decision tree (J48) + cluster-based kNN undersampling	55.68	68.50	54.86	0.6130	0.6020
One-sided selection + OB1	94.05	0	100	0	0.4997
One-sided selection + OB2	94.05	0	100	0	0.5000
GA-SVM + Rand undersampling with OB1	63.22	66.80	62.99	0.6487	0.7071
GA-SVM + Rand undersampling with OB2	62.92	67.64	62.62	0.6508	0.6885
GA-SVM + cluster-based kNN undersampling with OB1	65.72	65.12	65.76	0.6544	0.6997
GA-SVM + cluster-based kNN undersampling with OB2	62.67	65.96	62.46	0.6419	0.6599