ThalPred: a web-based prediction tool for discriminating thalassemia trait and iron deficiency anemia

Table 4 Performance comparisons between DT, RF and SVM models in differentiation of iron deficiency anemia from thalassemia trait

Classifier	Validation	Ac (%)	Sp (%)	Sn (%)	MCC	YI	AUC
k-NN	5-fold CV	92.36 ± 1.67	90.48 ± 3.62	92.80 ± 1.76	0.76 ± 0.06	0.83 ± 0.04	0.81 ± 0.07
	External	92.54 ± 4.26	90.09 ± 10.79	93.35 ± 4.22	0.77 ± 0.14	0.83 ± 0.13	0.80 ± 0.10
	Independent	90.20	83.33	100.00	0.82	0.83	0.85
DT	5-fold CV	98.03 ± 0.91	96.15 ± 3.40	98.54 ± 0.99	0.94 ± 0.026	0.93 ± 0.06	1.00 ± 0.00
	External	93.83 ± 4.10	86.86 ± 11.16	96.22 ± 3.77	0.82 ± 0.12	0.82 ± 0.04	0.92 ± 0.06
	Independent	92.16	86.21	100.00	0.85	0.85	1.00
RF	5-fold CV	94.17 ± 1.26	88.15 ± 3.49	95.75 ± 0.95	0.83 ± 0.04	0.83 ± 0.04	0.97 ± 0.01
	External	94.62 ± 3.29	90.07 ± 8.99	96.13 ± 3.19	0.84 ± 0.10	0.84 ± 0.07	0.98 ± 0.02
	Independent	92.16	86.21	100.00	0.85	0.85	1.00
ANN	5-fold CV	94.11 ± 1.31	86.75 ± 3.58	96.14 ± 1.06	0.83 ± 0.04	0.83 ± 0.04	0.97 ± 0.02
	External	93.78 ± 3.71	86.81 ± 10.84	96.22 ± 3.30	0.82 ± 0.11	0.83 ± 0.11	0.98 ± 0.02
	Independent	94.12	89.29	100.00	0.89	0.89	1.00
SVM	5-fold CV	95.05 ± 1.06	89.81 ± 2.66	96.45 ± 0.96	0.85 ± 0.03	0.85 ± 0.04	0.97 ± 0.01
	External	95.59 ± 2.76	92.49 ± 8.47	96.74 ± 2.59	0.87 ± 0.08	0.87 ± 0.10	0.98 ± 0.03
	Independent	96.08	92.59	100.00	0.92	0.92	1.00

The data are shown as mean ± standard deviation (100 times)
Ac Accuracy; ANN Artificial neural network; AUC Area under receiver operating curve; DT Decision tree; k-NN k-nearest neighbor; MCC Matthew’s correlation coefficient; RF Random forest; Sn Sensitivity; Sp Specificity; SVM Support vector machine; YI Youden’s index; 5-fold CV 5-fold cross validation
Parameters of k-NN (k), RF (ntree, mtry), ANN (size, decay) and SVM (cost, γ) were optimized by a 5-fold CV procedure. Values of k, ntree, mtry, size, decay, cost and γ are 5, 200, 2, 4, 0.5, 8 and 0.5

ISSN: 1472-6947