Galaxy Morphology Classification
An interpretable deep learning project for classifying galaxy morphologies (Elliptical vs Spiral) using the Galaxy Zoo 2 dataset from SDSS images. Achieved 99.07% accuracy and ROC-AUC 0.9995 with ResNet18 transfer learning. Features Grad-CAM and Integrated Gradients for scientific interpretability.
Proof points
- 99.07% Accuracy
- ROC-AUC 0.9995
Technologies
PyTorch, ResNet18, Grad-CAM, Captum, Astropy, NumPy
Links
Problem
Astronomical datasets require interpretable ML models where researchers can understand why predictions are made - critical for building trust in astrophysics research.
Approach
Used ResNet18 with ImageNet transfer learning on 24,273 high-confidence galaxy images (80%+ citizen science vote agreement). Implemented Grad-CAM and Integrated Gradients for visual interpretability.
Results
Achieved 99.07% test accuracy with near-perfect ROC-AUC (0.9995). Model attention aligns with expected morphological features - bulge regions for ellipticals, spiral arms for spirals.