Feature Attribution on Synthetic half-moons dataset with Gaussian noises (std dev 0.05-0.65)
0.531AUC-PurityGeodesic IG (kNN)
Evaluation results
| Method | Links | |
|---|---|---|
| Geodesic IG (kNN)baseline=(-0.5, -0.5), number of neighbors=15, variant=kNN2025.02 | 0.531 | |
| Occlusionbaseline=(-0.5, -0.5)2025.02 | 0.52 | |
| Geodesic IG (SVI)baseline=(-0.5, -0.5), variant=SVI2025.02 | 0.504 | |
| IGbaseline=(-0.5, -0.5)2025.02 | 0.487 | |
| GradientShapbaseline=(-0.5, -0.5)2025.02 | 0.483 | |
| Kernel Shapbaseline=(-0.5, -0.5)2025.02 | 0.48 | |
| Enhanced IGbaseline=(-0.5, -0.5), number of neighbors=152025.02 | 0.47 | |
| Guided IGbaseline=(-0.5, -0.5)2025.02 | 0.361 | |
| Input X Gradientsbaseline=(-0.5, -0.5)2025.02 | 0.328 | |
| Randombaseline=(-0.5, -0.5)2025.02 | 0.299 |