# Verified Results

- **Selected model:** `random_forest`
- **Usable objects:** 15,000
- **Training rows:** 12,000
- **Reserved holdout rows:** 3,000
- **Holdout macro F1:** 0.927
- **Holdout balanced accuracy:** 0.927
- **Holdout accuracy:** 0.927

## Per-class holdout performance

| Class | Precision | Recall | F1 | Support |
|---|---:|---:|---:|---:|
| STAR | 0.937 | 0.923 | 0.930 | 1,000 |
| GALAXY | 0.951 | 0.929 | 0.940 | 1,000 |
| QSO | 0.896 | 0.930 | 0.913 | 1,000 |

These values describe a same-survey holdout from a deliberately balanced,
spectroscopically targeted SDSS sample. They are not full-sky deployment metrics.