BERT Clinical Notes Multiclassification

This project fine-tunes Bio_Clinical BERT for multi-class classification of clinical notes using transformer-based deep learning.

Features

• Full pipeline: JSON data loading, preprocessing, train–validation stratified split, and label encoding
• Custom PyTorch Dataset class for integration with HuggingFace Trainer
• Fine-tunes pretrained Bio_ClinicalBERT for 22 clinical note categories
• Model evaluation with classification report, macro F1/accuracy, and annotated confusion matrix
• Visualization of per-class distribution and prediction results

Checking Class Imbalance

Stack Used

• Python, pandas, numpy
• PyTorch, HuggingFace Transformers, Datasets
• scikit-learn (evaluation, splitting)
• matplotlib, seaborn