Multi-Label Classification of Legal Cases According to the Sustainable Development Goals Using Machine Learning Algorithms

Multi-label text classification plays a crucial role in various domains, enabling the automated categorization of documents into multiple categories simultaneously. This study explores the application of multi-label text classification techniques to a dataset of 30,000 legal documents in Brazilian Portuguese, specifically judicial decisions from the Court of Justice of Rio Grande do Norte. Our research focuses on leveraging problem transformation methods, including Binary Relevance and Classifier Chains, and comparing the performance of four different classifiers: Random Forest, XGBoost, Support Vector Machine (SVM), and CatBoost. We utilize TFIDF to generate embeddings from the textual data and evaluate the classifiers' performance using metrics such as accuracy, Hamming Loss, and Zero-One Loss. Our findings indicate that XGBoost, particularly when combined with problem transformation methods, consistently outperforms other classifiers in terms of accuracy and robustness. This study provides valuable insights into the application of machine learning for classifying legal documents and contributes to the growing body of research on utilizing technology to support the achievement of the United Nations' Sustainable Development Goals (SDGs).