Classificação textual com uso de florestas aleatórias em processos jurídicos

Abstract

The judicial system undergoes constant transformations when analyzing issues of law, codes, and decrees. Technological innovation is also necessary in the legal field with proposals to optimize tasks such as case law search, predictive analysis of decisions, and process screening. This study investigated the application of text classification techniques in procedural documents in the last appeal stage to evaluate their effectiveness in manual labeling at the beginning of proceedings in second-degree jurisdiction. The performance metrics from the Random Forest Classifier were analyzed, revealing an average micro F1-score of 86,3%. Variations in metrics were observed for each class, highlighting the method´s performance in identifying the class “Drug Trafficking and Related Conduct” (TD). The removal of the TD class resulted in changes to the performance, highlighting the importance of the class structure in the implementation model practice. These results indicate that text classification techniques presented satisfactory results in analyzing procedural documents in the last appeal stage. However, it is essential to adapt the model to the particularities of the legal context to optimize its practical applicability.