Table 4 DC characterization results for different classification methods. All methods had a very small accuracy difference within 2% compared to the proposed method. Statistically significant differences (p < 0.05) compared with ‘RF’, ‘Adaboost’, ‘SVM’, ‘FFNN’, and ‘RBFNN’ are indicated by ‘*’, ‘γ’, ‘†’, ‘ψ’, and ‘δ’, respectively, as determined from the student’s t-test. Data are shown as mean ± standard deviation
Methods | PPV (%) | NPV (%) | Sensitivity (%) | Specificity (%) | Accuracy (%) | AUC |
|---|---|---|---|---|---|---|
RF | 87.1 ± 0.1 | 87.7 ± 0.1 | 87.1 ± 0.1 | 87.8 ± 0.1 | 87.4 ± 0.1 | 0.874 ± 0.003 |
Adaboost | 87.1 ± 0.1 | 86.3 ± 0.1 | 87.3 ± 0.2 | 86.1 ± 0.1 | 86.7 ± 0.1 | 0.867 ± 0.002 |
SVM | 87.7 ± 0.1 | 89.8 ± 0.1 | 88.0 ± 0.0 | 89.6 ± 0.0 | 88.8 ± 0.1 | 0.888 ± 0.001 |
FFNN | 83.6 ± 0.2 | 93.3 ± 0.1 | 94.2 ± 0.3 | 81.4 ± 0.1 | 87.8 ± 0.2 | 0.885 ± 0.001 |
RBFNN | 85.9 ± 0.1 | 89.8 ± 0.1 | 85.1 ± 0.1 | 90.3 ± 0.1 | 87.7 ± 0.1 | 0.878 ± 0.001 |
Proposed Method | 86.0 ± 0.1ψ | 91.2 ± 0.1*γ | 91.8 ± 0.1*γ†δ | 85.1 ± 0.1†ψδ | 88.4 ± 0.1 | 0.886 ± 0.001 |