International Journal of Dentistry

Research Article

Identification of Microbial and Proteomic Biomarkers in Early Childhood Caries

Table 2

Performance statistics of three classification models tested on the MS proteomics data. The models were optimized for the average misclassification error (zero-one loss). Four different MS datasets generated for combinations of two affinity chips (CM-10 and Q-10) and two intensity instrument settings (high and low) were analyzed. The statistics include averages and standard deviations of test errors, sensitivities, and specificities of respective classifiers. The averages and standard deviations were calculated across 40 different train/test obtained through the random subsampling approach.
“caries cm 10 high”Test errorSensitivitySpecificity
SVM 3 1 . 8 2 % ± 5 . 3 5 % 6 6 . 1 0 % ± 7 . 4 9 % 6 9 . 8 0 % ± 9 . 4 4 %
SVM 100 WLCX 3 5 . 6 8 % ± 5 . 5 8 % 5 7 . 7 2 % ± 2 1 . 5 6 % 7 0 . 9 3 % ± 1 8 . 7 2 %
Rnd Forest 3 2 . 9 5 % ± 5 . 7 4 % 5 4 . 2 4 % ± 1 2 . 2 9 % 7 9 . 1 7 % ± 1 0 . 8 4 %
“caries cm10 low”Test errorSensitivitySpecificity
SVM 2 8 . 2 3 % ± 5 . 8 2 % 6 9 . 8 3 % ± 8 . 1 9 % 7 3 . 8 1 % ± 9 . 6 2 %
SVM 100 WLCX 2 6 . 6 8 % ± 5 . 7 8 % 7 3 . 6 4 % ± 1 1 . 6 6 % 7 3 . 5 5 % ± 1 0 . 6 6 %
Rnd Forest 2 5 . 6 4 % ± 5 . 7 6 % 6 5 . 2 9 % ± 9 . 5 5 % 8 3 . 2 0 % ± 9 . 5 4 %
“caries q10 high”Test errorSensitivitySpecificity
SVM 2 5 . 9 1 % ± 4 . 8 8 % 7 3 . 3 1 % ± 6 . 8 7 % 7 5 . 0 5 % ± 8 . 1 4 %
SVM 100 WLCX 2 5 . 9 1 % ± 4 . 8 8 % 7 3 . 3 1 % ± 6 . 8 7 % 7 5 . 0 5 % ± 8 . 1 4 %
Rnd Forest 3 2 . 0 0 % ± 4 . 4 7 % 5 7 . 4 1 % ± 1 1 . 2 1 % 7 8 . 3 1 % ± 9 . 7 0 %
“caries q10 low”Test errorSensitivitySpecificity
SVM 2 2 . 7 3 % ± 3 . 9 3 % 7 5 . 8 2 % ± 9 . 6 2 % 7 8 . 8 8 % ± 6 . 9 4 %
SVM 100 WLCX 2 6 . 1 4 % ± 4 . 8 5 % 7 1 . 9 1 % ± 1 3 . 0 0 % 7 5 . 4 5 % ± 9 . 0 7 %
Rnd Forest 2 5 . 5 0 % ± 5 . 6 4 % 6 9 . 3 9 % ± 1 1 . 2 3 % 7 9 . 9 9 % ± 9 . 3 4 %
SVM: linear support vector machine.
SVM on the top 100 Wilcoxon peaks.
Random forest.