Table 1

Comparison of different classification methods

Recognition rate

Prediction rate

Sensitivity

Specificity

Accuracy rate


Ctrl/PLS-DA

95.5% (63/66)

75.0% (24/32)

91.2% (52/57)

85.4% (35/41)

89.8% (88/98)

UV/PLS-DA

98.5% (65/66)

78.1% (25/32)

91.2% (52/57)

92.7% (38/41)

91.8% (90/98)

DOSC/PLS-DA

100% (66/66)

84.4% (27/32)

93.0% (53/57)

95.1% (40/41)

94.9% (93/98)

O-PLS/PLS-DA

100% (66/66)

81.3% (26/32)

91.2% (52/57)

95.1% (40/41)

93.9% (92/98)

FC/DOSC/PLS-DA

98.5% (65/66)

90.6% (29/32)

94.7% (54/57)

97.6% (40/41)

95.9% (94/98)

KNN (K = 3)

95.5% (63/66)

71.9% (23/32)

84.2% (48/57)

92.7% (38/41)

87.8% (86/98)

SIMCA

90.9% (60/66)

75.0% (24/32)

87.7% (50/57)

82.9% (34/41)

85.7% (84/98)

FC/KNN (K = 3)

95.5% (63/66)

81.3% (26/32)

93.0% (53/57)

87.8% (36/41)

90.8% (89/98)

SVM

100% (66/66)

81.3% (26/32)

94.7% (54/57)

92.7% (38/41)

93.9% (92/98)

DOSC/SVM

100% (66/66)

87.5% (28/32)

94.7% (54/57)

97.6% (40/41)

95.9% (94/98)

FC/SVM

100% (66/66)

90.6% (29/32)

96.5% (55/57)

97.6% (40/41)

96.9% (95/98)

FC/DOSC/SVM

100% (66/66)

96.9% (31/32)

100% (57/57)

97.6% (40/41)

99.0% (97/98)


Prediction results were from different classification methods (25 healthy and 41 diabetic samples in the training set; 16 healthy and 16 diabetic samples in the testing set). Recognition rate is the correctly classified rate in the training set. Prediction rate is the correctly classified rate in the testing set. Sensitivity is the rate of true positive classified as positive. Specificity is the rate of true negative classified as negative (Ctrl, mean-centered scaling; UV, auto scaling; DOSC, direct orthogonal signal correction; O-PLS, orthogonal projections to latent structures; FC, Fisher's criterion for feature selection).

Wang et al. BMC Bioinformatics 2009 10:83   doi:10.1186/1471-2105-10-83

Open Data