Urdu-English Code-Switching Classification Model
This model is an XLM-RoBERTa-based token classification model trained for Urdu-English code-switching detection.
Labels
- URD — Urdu
- ENG — English
- MIX — Mixed
Training
- Model: xlm-roberta-base
- Epochs: 5
- Dataset size: 971 rows
Evaluation
| Metric | Score |
|---|
| Accuracy | 96.43% |
| Precision | 96.64% |
| Recall | 96.43% |
| Weighted F1 | 96.12% |
F1 by Label
| Label | F1 |
|---|
| URD | 0.99 |
| ENG | 0.95 |
| MIX | 0.71 |
Confusion Matrix
The model performs strongly on URD and ENG. MIX has lower recall because the test set contained only 11 MIX examples.
