Using feature engineering to improve the identification of bias in legal judgment database

• Autor(a) principal: Carvalho, Daniel Marx Pinto
• Data de Publicação: 2022
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: eng
• Título da fonte: Repositório Institucional da UFRN
• Texto Completo: https://repositorio.ufrn.br/handle/123456789/48478
• topic: Machine Learning; Feature Selection; Undersampling; Prediction of bias in court decisions

Machine learning techniques have been widely applied to judicial data. The use of thesetechniques allows the extraction of patterns that exist among the data, which can beused, for example, to predict the outcome of court decisions. In Brazil, the use of machine learning in the legal area is already a reality, having been enabled by the work of digitizing legal processes, which has been taking place since the beginning of the century. In general, judicial databases are characterized by high dimension. If the training dataset has irrelevant or redundant attributes, or even a large variation in the sample size of the classes you want to classify (that is, an unbalanced database), the results of the classification analysis step may produce less accurate results. In addition to that, a large number of attributes increases the computational power required for the task and the complexity of the classification models used. In these circumstances, the data pre-processing step is essential to prepare the database for analysis and increase the accuracy of predictions. Judicial data has yet another important characteristic, its collection, and identification of features often require specialized professionals in the legal area and a manual process of building the database, which makes it slow, not to mention that there may be incomplete data on a particular feature. The objective of the work in question is, therefore, to make use feature engineering techniques, such as selection and undersampling, on a dataset of judicial sentences (built by the research group ’Além da Pena’), in order to investigate the detection of gender biases in the judicial decisions present in the base.