Scholarlink

The use of Information and Communication Technologies (ICT) in the justice sector has been inconsistent globally, influenced by political decisions and the specific characteristics of each legal system. In this context, the implementation of predictive litigation models emerges as one of the most effective applications for improving efficiency in the defense and resolution of judicial cases, regardless of the legal context in which they are applied. Therefore, this study explores the application of machine learning techniques, Random Forest (RF) and Gradient Boosting (GB), to predict judicial rulings using data from the Peruvian Constitutional Court, specifically decisions issued between 2021 and 2024. The results indicate that GB slightly outperforms RF in terms of accuracy and F1-Score, positioning it as the preferred technique. However, limitations were observed in predicting minority classes, as revealed by confusion matrices and performance metrics. For future research, it is recommended to explore methods such as SMOTE to enhance accuracy in more complex judicial scenarios.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluation of Predictive Models in the Justice Sector: A Comparison Between Random Forest and Gradient Boosting

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Renato Arias,
  • Ricardo Arias,
  • Kelly Ochoa

摘要

The use of Information and Communication Technologies (ICT) in the justice sector has been inconsistent globally, influenced by political decisions and the specific characteristics of each legal system. In this context, the implementation of predictive litigation models emerges as one of the most effective applications for improving efficiency in the defense and resolution of judicial cases, regardless of the legal context in which they are applied. Therefore, this study explores the application of machine learning techniques, Random Forest (RF) and Gradient Boosting (GB), to predict judicial rulings using data from the Peruvian Constitutional Court, specifically decisions issued between 2021 and 2024. The results indicate that GB slightly outperforms RF in terms of accuracy and F1-Score, positioning it as the preferred technique. However, limitations were observed in predicting minority classes, as revealed by confusion matrices and performance metrics. For future research, it is recommended to explore methods such as SMOTE to enhance accuracy in more complex judicial scenarios.