Table 2.
Accuracy of the classification algorithm with subsequent reduction in included m/z values.
| RF Model | #1 | #2 | #3 | #4 | #5 | #6 | #7 | #8 | #9 |
|---|---|---|---|---|---|---|---|---|---|
| Included variables (m/z values) |
all m/z values (n = 195) | 1410.7 810.4 1406.7 865.4 878.5 1234.7 1220.7 1104.6 |
1410.7 810.4 1406.7 865.4 878.5 1234.7 1220.7 |
1410.7 810.4 1406.7 865.4 878.5 1234.7 |
1410.7 810.4 1406.7 865.4 878.5 |
1410.7 810.4 1406.7 865.4 |
1410.7 810.4 1406.7 |
1410.7 810.4 |
1410.7 |
| Model Metrics | |||||||||
| Prediction accuracy (CI95%) | 0.906 (0.868–0.935) |
0.906 (0.868–0.935) |
0.903 (0.865–0.9328) |
0.912 (0.875–0.941) |
0.887 (0.847–0.919) |
0.884 (0.843–0.917) |
0.887 (0.847–0.919) |
0.726 (0.674–0.775) |
0.676 (0.622–0.727) |
| Sensitivity | 0.947 | 0.929 | 0.929 | 0.929 | 0.894 | 0.882 | 0.894 | 0.781 | 0.686 |
| Specificity | 0.859 | 0.879 | 0.873 | 0.893 | 0.879 | 0.886 | 0.879 | 0.664 | 0.664 |
| PPV | 0.884 | 0.897 | 0.892 | 0.908 | 0.894 | 0.898 | 0.894 | 0.725 | 0.699 |
| NPV | 0.934 | 0.916 | 0.916 | 0.917 | 0.879 | 0.868 | 0.879 | 0.728 | 0.651 |
| Validation Data Set: n =318 (ADC n = 169, SqCC n = 149) | |||||||||
| Misclassifications, n (%) | |||||||||
| ADC misclassified as SqCC | 9 (5) | 12 (7) | 12 (7) | 12 (7) | 18 (11) | 20 (12) | 18 (11) | 37 (22) | 53 (31) |
| SqCC misclassified as ADC | 21 (14) | 18 (12) | 19 (13) | 16 (11) | 18 (12) | 17 (11) | 18 (12) | 50 (34) | 50 (34) |
| Overall misclassified | 30 (9) | 30 (9) | 31 (10) | 28 (9) | 36 (11) | 37 (12) | 36 (11) | 87 (27) | 103 (32) |
ADC, adenocarcinoma; NPV, negative predictive value; PPV, positive predictive value; RF, random forest; SqCC, squamous cell carcinoma.