Predicting applicable law sections from judicial case reports using legislative text analysis with machine learning

This paper presents a study on legislative text analysis to automate the process of identifying appropriate sections of laws that are applicable to the cases. We propose a methodology that includes supervised machine learning (ML) and natural language processing (NLP), and demonstrated our idea on the archived case studies of Indian Income Tax Act of 1963 (Income tax act, 1961 complete act—bare act, 2008), with applicable law sections and subsections, available at ‘LegalCrystal’ ( https://www.legalcrystal.com/ ) data repository. We consider the problem as a multi-label classification task, where multiple law sections could be applied on one case. The one-versus-rest wrapper is applied over the conventional ML models like logistic regression, Naïve bayes, decision tree and support vector machine to perform the multi-label classification. The proposed methodology includes necessary preprocessing and word embedding of texts, pipelining of transformers and ML models and evaluation of the trained models. We analyzed the performance of these different ML models by fine-tuning the hyper-parameters and observed a highest F1 score of 0.75 for support vector machine. Although this work is limited to cases involving income tax laws, the proposed methodology is adaptive to any other law sections.