Scholarlink

<p>In classification problems involving financial decision making, cost-sensitive classifiers have been widely used to minimize economic losses, which are often evaluated through specific metrics such as the savings score. However, these methods frequently compromise conventional performance metrics such as accuracy, precision, and recall. In this light, this work presents R3D, a hybrid classification approach that integrates traditional and example-dependent cost-sensitive classifiers to improve the classification of revision requests in a Brazilian tax administration service. This integration between those classifiers aims to take advantage of their complementary characteristics. In public administration, particularly in tax-related services, achieving a balance between cost reduction and decision reliability is essential. By defining a threshold based on debt values, the proposed method ensures that high-cost misclassifications are minimized without excessively increasing error rates in lower-cost cases. Experimental results show that R3D outperforms both purely cost-sensitive and traditional classifiers by achieving a more balanced trade-off between cost efficiency and predictive performance, making it a robust solution for government decision-making processes.</p>

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

R3D: an approach combining cost-sensitive and insensitive classifiers to achieve more balanced scores

  • 期刊论文

  • 发表日期：

    2025年6月27日

查看全文
  • Helton Souza Lima,
  • Damires Yluska de Souza Fernandes,
  • Thiago José Marques Moura

摘要

In classification problems involving financial decision making, cost-sensitive classifiers have been widely used to minimize economic losses, which are often evaluated through specific metrics such as the savings score. However, these methods frequently compromise conventional performance metrics such as accuracy, precision, and recall. In this light, this work presents R3D, a hybrid classification approach that integrates traditional and example-dependent cost-sensitive classifiers to improve the classification of revision requests in a Brazilian tax administration service. This integration between those classifiers aims to take advantage of their complementary characteristics. In public administration, particularly in tax-related services, achieving a balance between cost reduction and decision reliability is essential. By defining a threshold based on debt values, the proposed method ensures that high-cost misclassifications are minimized without excessively increasing error rates in lower-cost cases. Experimental results show that R3D outperforms both purely cost-sensitive and traditional classifiers by achieving a more balanced trade-off between cost efficiency and predictive performance, making it a robust solution for government decision-making processes.