TABLE 3.
Results of the five average metrics scores from 50 different train–test-split experiments (mean ± SD) on the TCGA lung cancer data set. The accuracy, precision, recall, and f1-score were calculated with the optimal threshold selected using Youden’s J statistic.
| Models | Average scores of 50 experiments on TCGA datasets | ||||
|---|---|---|---|---|---|
| AUC | Accuracy | Precision | Recall | F1-score | |
| DL-four-inputs | 71.48 ± 4% | 69.85 ± 6% | 69.17 ± 11% | 87.93 ± 4% | 76.66 ± 6% |
| DL-three-inputs-age | 65.68 ± 4% | 64.42 ± 8% | 62.34 ± 15% | 86.39 ± 4% | 71.00 ± 10% |
| DL-three-inputs-stage | 70.69 ± 4% | 68.95 ± 7% | 68.29 ± 14% | 87.54 ± 4% | 75.64 ± 8% |
| DL-two-inputs | 65.16 ± 4% | 62.82 ± 9% | 59.31 ± 17% | 87.22 ± 5% | 68.72 ± 11% |
| DL-one-input-BRITE | 63.58 ± 4% | 62.74 ± 9% | 61.03 ± 17% | 85.13 ± 4% | 69.32 ± 11% |
| DL-one-input-pathway | 64.69 ± 4% | 63.31 ± 8% | 60.97 ± 17% | 86.32 ± 5% | 69.62 ± 11% |
| KNN | 53.63 ± 5% | 57.22 ± 11% | 52.51 ± 23% | 85.47 ± 6% | 61.54 ± 16% |
| SVM | 54.77 ± 5% | 56.11 ± 11% | 52.69 ± 23% | 84.17 ± 7% | 60.58 ± 18% |
| Random-forest | 57.41 ± 6% | 57.33 ± 12% | 53.40 ± 24% | 85.09 ± 7% | 61.68 ± 18% |
| Logistic-regression | 50.81 ± 5% | 55.41 ± 15% | 53.91 ± 29% | 82.50 ± 8% | 58.67 ± 25% |
| MLP | 55.06 ± 5% | 54.61 ± 11% | 49.14 ± 21% | 83.91 ± 5% | 58.75 ± 17% |
The bold values are the highest among all the models.