Table 13 Comparison of Off‑policy PPO FS with classical, advanced, and RL-based methods on the CelebA dataset.
Model | Accuracy | F-measure | G-means | AUC |
|---|---|---|---|---|
LASSO | 68.511 ± 0.027 | 70.211 ± 0.048 | 71.881 ± 0.041 | 0.613 ± 0.013 |
mRMR | 69.609 ± 0.074 | 71.678 ± 0.082 | 72.314 ± 0.085 | 0.628 ± 0.086 |
PCA | 67.511 ± 0.043 | 69.202 ± 0.019 | 71.929 ± 0.070 | 0.603 ± 0.080 |
MI | 70.772 ± 0.049 | 72.270 ± 0.066 | 73.909 ± 0.088 | 0.632 ± 0.016 |
Cross-Attention | 71.582 ± 0.074 | 74.345 ± 0.005 | 75.995 ± 0.084 | 0.648 ± 0.062 |
AE | 72.948 ± 0.012 | 74.437 ± 0.005 | 75.138 ± 0.085 | 0.652 ± 0.066 |
TSFS | 74.223 ± 0.058 | 76.531 ± 0.072 | 77.186 ± 0.069 | 0.658 ± 0.088 |
A-SFS | 75.209 ± 0.089 | 77.730 ± 0.062 | 78.395 ± 0.073 | 0.661 ± 0.010 |
FaceNet | 75.987 ± 0.077 | 78.495 ± 0.072 | 79.157 ± 0.082 | 0.667 ± 0.059 |
VGG-Face | 77.220 ± 0.003 | 78.351 ± 0.018 | 79.048 ± 0.078 | 0.681 ± 0.070 |
DeepFace | 77.810 ± 0.063 | 79.956 ± 0.026 | 80.655 ± 0.062 | 0.689 ± 0.083 |
RL | 80.860 ± 0.054 | 82.556 ± 0.029 | 83.198 ± 0.090 | 0.782 ± 0.023 |
SAC | 83.579 ± 0.076 | 84.102 ± 0.012 | 84.631 ± 0.026 | 0.798 ± 0.053 |
PPO | 81.991 ± 0.052 | 82.887 ± 0.067 | 83.449 ± 0.069 | 0.792 ± 0.095 |
Off-policy PPO | 87.951 ± 0.096 | 89.409 ± 0.087 | 90.193 ± 0.030 | 0.829 ± 0.073 |
