Results
| Model | Classification Accuracy | Area under ROC curve |
|---|---|---|
| CNN | 94.72% | 0.94 |
| XGBoost + PCA | 86.75% | 0.85 |
Transfer Learning (ResNet18_2) |
96.25% | 0.96 |
Transfer Learning (ResNet34) |
94.75% | 0.95 |
Transfer Learning (ResNet18) |
92.89% | 0.93 |
Transfer Learning (Xception) |
79.19% | 0.77 |
| CNN Ensemble | 93.77% | 0.93 |
| XGBoost | 89.12% | 0.95 |
| FFT+XGBoost | 83.79% | <not_yet_calc> |
| SMOTE resampling | 89.29% | 0.95 |
Some statistics of the best model (Resnet18_2) are 1:
| Recall | Precision | F1 score |
|---|---|---|
| 0.966 | 0.9695 | 0.9677 |
[1] These values were calculated after a single experiment but could change slightly on a different pass.
References
Ackermann, S., Schawinski, K., Zhang, C., Weigel, A., & Turp, M. (2018). Using transfer learning to detect galaxy mergers. Monthly Notices Of The Royal Astronomical Society, 479(1), 415-425. doi: 10.1093/mnras/sty1398