Table 1.
Performance of confidence methods for colon cancer (62 samples: 22 class 1, 40 class 2)a.
| LEVEL | N | POS | ACC | SEN | SPC | PPV | NPV |
|---|---|---|---|---|---|---|---|
| Overall | 61.0(0.9) | 65.1(1.3) | 90.2(1.4) | 91.9(1.0) | 86.9(3.3) | 92.9(1.8) | 85.2(2.0) |
| Method 1 | |||||||
| Highest | 20.0(2.4) | 82.7(4.3) | 94.7(2.7) | 96.3(3.2) | 86.8(16.2) | 97.4(3.0) | 84.5(13.8) |
| Intermediate | 36.3(3.0) | 60.0(3.6) | 89.5(2.2) | 90.8(4.0) | 87.5(4.8) | 91.7(2.7) | 86.7(4.9) |
| Lowest | 4.8(1.7) | 30.3(17.1) | 76.9(18.3) | 52.3(42.6) | 85.1(18.4) | 59.4(42.1) | 83.3(16.5) |
| Method 2 | |||||||
| Highest | 51.5(2.6) | 70.9(1.5) | 92.0(1.5) | 94.1(1.9) | 86.4(4.9) | 94.5(1.7) | 86.0(4.1) |
| Intermediate | 4.9(1.0) | 26.2(21.2) | 88.5(12.7) | 78.9(31.2) | 90.7(15.6) | 78.9(31.2) | 92.8(12.0) |
| Lowest | 4.7(1.8) | 38.6(19.3) | 74.6(20.8) | 57.8(36.9) | 81.3(23.4) | 70.2(25.5) | 79.1(21.3) |
The first row shows the overall performance of the selective-voting algorithm based on 20 repetitions of 10-fold CV. Column 1 designates the subpopulation confidence, where Method 1 corresponds to ‘clustering training samples according to AM and assigning test samples according to VM’, and Method 2 corresponds to ‘clustering training samples according to AV and assigning test samples according to VV’ (see section 2.2). N (column 2) is the average number of samples in each subpopulation among the total samples that were classified by the selective-voting algorithm across the 20 repetitions, and POS (column 3) indicates the average percentage of positive (class 2) samples among the N samples. The performance measures in columns 4–8 are defined in section 2.4. The numbers in parentheses are standard deviations.