Table 1.
Results of LDA and linear SVM (linSVM) classification experiments based on local deformations reduced by pLDA with resampling using various values of λ, PCA, t-test with three thresholds or unreduced.
| Reduction | λ | Classifier | LOOCV during classification—LOOCV(c) | LOOCV during reduction and classification—LOOCV(rc) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| # features | # iterations (mean±SD) | Acc (%) | Sens (%) | Spec (%) | # features (mean±SD) | # iterations (mean±SD) | Acc (%) | Sens (%) | Spec (%) | |||
| pLDA | 0.6 | LDA | 98,320 | 3.0 ± 0.0 | 84.6 | 86.5 | 82.7 | 104,184.2 ± 8527.3 | 3.0 ± 0.0 | 65.4 | 57.7 | 73.1 |
| linSVM | 86.5 | 88.5 | 84.6 | 69.2 | 67.3 | 71.2 | ||||||
| pLDA | 0.7 | LDA | 83,193 | 3.0 ± 0.0 | 83.7 | 84.6 | 82.7 | 89,008.5 ± 7796.1 | 3.0 ± 0.0 | 65.4 | 57.7 | 73.1 |
| linSVM | 86.5 | 88.5 | 84.6 | 70.2 | 67.3 | 73.1 | ||||||
| pLDA | 0.8 | LDA | 71,245 | 3.0 ± 0.1 | 82.7 | 82.7 | 82.7 | 76,965.1 ± 7170.8 | 3.0 ± 0.0 | 65.4 | 57.7 | 73.1 |
| linSVM | 86.5 | 88.5 | 84.6 | 70.2 | 65.4 | 75.0 | ||||||
| pLDA | 0.9 | LDA | 61,398 | 2.9 ± 0.3 | 82.7 | 82.7 | 82.7 | 66,372.6 ± 5736.2 | 3.0 ± 0.2 | 66.3 | 59.6 | 73.1 |
| linSVM | 86.5 | 88.5 | 84.6 | 64.4 | 63.5 | 65.4 | ||||||
| pLDA | 1.0 | LDA | 52,927 | 2.9 ± 0.3 | 81.7 | 80.8 | 82.7 | 58,355.1 ± 5250.0 | 2.9 ± 0.3 | 67.3 | 61.5 | 73.1 |
| linSVM | 85.6 | 86.5 | 84.6 | 68.3 | 65.4 | 71.2 | ||||||
| pLDA | 1.1 | LDA | 45,557 | 3.1 ± 0.5 | 82.7 | 80.8 | 84.6 | 50,510.9 ± 4896.3 | 3.0 ± 0.4 | 66.3 | 59.6 | 73.1 |
| linSVM | 84.6 | 82.7 | 86.5 | 63.5 | 61.5 | 65.4 | ||||||
| pLDA | 1.2 | LDA | 39,008 | 3.2 ± 0.6 | 83.7 | 82.7 | 84.6 | 44,624.4 ± 5221.4 | 3.0 ± 0.6 | 66.3 | 57.7 | 75.0 |
| linSVM | 84.6 | 82.7 | 86.5 | 66.3 | 65.4 | 67.3 | ||||||
| pLDA | 1.3 | LDA | 33,082 | 3.6 ± 0.7 | 84.6 | 82.7 | 86.5 | 38,712.8 ± 4816.2 | 3.3 ± 0.6 | 65.4 | 55.8 | 75.0 |
| linSVM | 84.6 | 82.7 | 86.5 | 67.3 | 65.4 | 69.2 | ||||||
| pLDA | 1.4 | LDA | 27,807 | 3.9 ± 0.8 | 84.6 | 82.7 | 86.5 | 33,401.8 ± 4425.2 | 3.7 ± 0.6 | 65.4 | 57.7 | 73.1 |
| linSVM | 82.7 | 80.8 | 84.6 | 66.3 | 65.4 | 67.3 | ||||||
| pLDA | 1.5 | LDA | 23,116 | 4.2 ± 0.8 | 85.6 | 82.7 | 88.5 | 28,275.0 ± 3403.0 | 4.1 ± 0.7 | 68.3 | 63.5 | 73.1 |
| linSVM | 81.7 | 80.8 | 82.7 | 63.5 | 61.5 | 65.4 | ||||||
| pLDA | 1.6 | LDA | 18,917 | 4.7 ± 0.9 | 85.6 | 82.7 | 88.5 | 24,092.3 ± 3688.9 | 4.4 ± 0.7 | 67.3 | 61.5 | 73.1 |
| linSVM | 81.7 | 80.8 | 82.7 | 68.3 | 67.3 | 69.2 | ||||||
| pLDA | 1.7 | LDA | 15,203 | 5.0 ± 1.1 | 84.6 | 80.8 | 88.5 | 19,575.5 ± 3001.0 | 4.6 ± 0.9 | 67.3 | 61.5 | 73.1 |
| linSVM | 81.7 | 80.8 | 82.7 | 63.5 | 63.5 | 63.5 | ||||||
| pLDA | 1.8 | LDA | 11,736 | 5.6 ± 1.5 | 85.6 | 80.8 | 90.4 | 16,241.5 ± 2996.4 | 5.0 ± 1.0 | 63.5 | 59.6 | 67.3 |
| linSVM | 80.8 | 78.8 | 82.7 | 63.5 | 65.4 | 61.5 | ||||||
| pLDA | 1.9 | LDA | 9555 | 7.1 ± 9.5 | 83.7 | 78.8 | 88.5 | 12,483.7 ± 2421.6 | 5.5 ± 1.2 | 65.4 | 61.5 | 69.2 |
| linSVM | 80.8 | 80.8 | 80.8 | 55.8 | 57.7 | 53.8 | ||||||
| pLDA | 2.0 | LDA | 7306 | 8.7 ± 13.2 | 83.7 | 78.8 | 88.5 | 9,910.8 ± 2196.8 | 6.2 ± 1.4 | 62.5 | 59.6 | 65.4 |
| linSVM | 80.8 | 78.8 | 82.7 | 59.6 | 59.6 | 59.6 | ||||||
| PCA | – | LDA | – | – | 60.6 | 55.8 | 65.4 | – | – | 66.3 | 65.4 | 67.3 |
| linSVM | 67.3 | 61.5 | 73.1 | 67.3 | 65.4 | 69.2 | ||||||
| t-test, p < 0.01 | – | LDA | 62,644 | – | 82.7 | 82.7 | 82.7 | 62,110.1 ± 3241.1 | – | 66.3 | 59.6 | 73.1 |
| linSVM | 84.6 | 84.6 | 84.6 | 70.2 | 69.2 | 71.2 | ||||||
| t-test, p < 0.005 | – | LDA | 38,630 | – | 84.6 | 82.7 | 86.5 | 38,184.7 ± 2389.9 | – | 65.4 | 57.7 | 73.1 |
| linSVM | 84.6 | 82.7 | 86.5 | 66.3 | 65.4 | 67.3 | ||||||
| t-test, p < 0.001 | – | LDA | 10,860 | – | 83.7 | 78.8 | 88.5 | 10,671.2 ± 1110.8 | – | 60.6 | 57.7 | 63.5 |
| linSVM | 78.8 | 76.9 | 80.8 | 59.6 | 63.5 | 55.8 | ||||||
| no | – | LDA | 1,924,670 | – | 67.3 | 63.5 | 71.2 | – | – | – | – | – |
| linSVM | 64.4 | 63.5 | 65.4 | – | – | – | ||||||
In columns: the number of selected features, the number of iterations of the pLDA algorithm and cross-validated classification performance measures in percentage—accuracy (Acc), sensitivity (Sens), and specificity (Spec).