TABLE 5.
Results of the five average metrics scores from 50 different train–test-split experiments (mean ± SD) on the GEO GSE37745 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 GEO datasets | ||||
|---|---|---|---|---|---|
| AUC | Accuracy | Precision | Recall | F1-score | |
| DL-four-inputs | 72.51 ± 6% | 73.85 ± 6% | 77.39 ± 14% | 79.26 ± 7% | 77.18 ± 7% |
| DL-three-inputs-age | 70.77 ± 5% | 71.03 ± 5% | 68.96 ± 17% | 81.26 ± 7% | 72.60 ± 9% |
| DL-three-inputs-stage | 72.36 ± 6% | 72.46 ± 6% | 71.04 ± 16% | 81.32 ± 7% | 74.39 ± 8% |
| DL-two-inputs | 69.74 ± 6% | 69.74 ± 6% | 65.30 ± 17% | 82.33 ± 9% | 70.58 ± 10% |
| DL-one-input-BRITE | 68.88 ± 5% | 70.56 ± 5% | 70.52 ± 14% | 79.10 ± 8% | 73.16 ± 7% |
| DL-one-input-pathway | 67.37 ± 5% | 68.05 ± 5% | 62.70 ± 15% | 80.91 ± 9% | 68.89 ± 8% |
| KNN | 55.76 ± 8% | 63.85 ± 9% | 56.35 ± 26% | 82.35 ± 13% | 60.84 ± 20% |
| SVM | 54.32 ± 8% | 61.33 ± 6% | 63.13 ± 23% | 72.04 ± 10% | 63.28 ± 15% |
| Random-forest | 55.59 ± 8% | 60.72 ± 7% | 52.78 ± 23% | 77.37 ± 11% | 58.21 ± 17% |
| Logistic-regression | 54.08 ± 8% | 58.51 ± 7% | 49.83 ± 24% | 75.82 ± 11% | 55.07 ± 17% |
| MLP | 54.69 ± 8% | 59.03 ± 7% | 49.04 ± 24% | 75.89 ± 9% | 55.56 ± 15% |
The bold values are the highest among all the models.