FIR Classification with BERT

This project uses a custom BERT model to classify synthetic FIR texts into crime categories.

Dataset

• Synthetic FIR data with labels like 'Assault', 'Fraud', 'Cybercrime', etc.
• Stored in data/fir_dataset.csv

Model

• Fine-tunes bert-base-uncased using HuggingFace's Trainer.

Setup

pip install -r requirements.txt
python train.py

Notes

• The dataset is synthetically generated and not from any real-world source.
• Accuracy is currently low — feel free to experiment and improve!

Notebook

You can also run the training pipeline interactively in fir_classifier_notebook.ipynb.