High-precision SMS Spam detection system using BERT. Achieved 98.0% accuracy through fine-tuning with PyTorch and Transformers.
More than 5500 real text labeled examples. Dataset created by Tiago A. Almeida and José M. Gómez Hidalgo available here.
- Data Visualization: Several charts show relevant data features.
- Transformer-based: Leverages
bert-base-casedfor deep contextual understanding. - Optimized Training: Implements a Linear Learning Rate Scheduler with a peak LR of 2e-5, decaying to zero to ensure stable convergence and prevent overfitting.
- High Performance: Reached 98.0% accuracy on the test set.
- Ready for Production: Model weights are hosted on Hugging Face for easy integration.
The model shows exceptional performance across all metrics:
| Metric | Score |
|---|---|
| Accuracy | 0.98 |
| Precision (Spam) | 0.99 |
| Recall (Spam) | 0.92 |
| F1-Score (Spam) | 0.93 |
Due to GitHub's file size limitations, the trained model weights are hosted on the Hugging Face Hub.
You can access the model here: [https://huggingface.co/mrcsgh/bert-sms-spam-classifier]