Multi-Label Legal Document Classification

ABSTRACT The rapid growth of digital legal documents has created a need for automated and accurate classification systems. Legal texts often belong to multiple overlapping legal domains, making single-label classification approaches inadequate. Traditional machine learning models struggle to capture the complex language and contextual structure of legal documents. This project proposes a Multi-Label Legal Document Classification System using Legal-BERT, a transformer model pretrained on large-scale legal corpora. The system preprocesses legal text, generates contextual embeddings using Legal-BERT, and applies a sigmoid-based classification layer to predict multiple legal categories simultaneously. Model performance is evaluated using precision, recall, and F1-score. A Streamlit-based web application enables real-time document classification with confidence scores. The results demonstrate improved accuracy and practical usability for automated legal document analysis. Keywords: Multi-Label Classification, Legal NLP, Legal-BERT, Transformer Models, Deep Learning, Streamlit, Legal Document Analysis