Scholarlink

This study compares subjective expert-based methods and data-driven predictive models for risk classification in Vehicle Protection Associations (VPAs). Using a dataset of 2,880 member records and fivefold cross-validation, Gradient Boosting achieved the best performance (86.59% accuracy, 82.50% recall), surpassing Logistic Regression, Artificial Neural Networks (ANNs), and the Analytic Hierarchy Process (AHP). SHAP analysis identified “Monthly Fee” and “FIPE Value” as key variables, contrasting with experts’ emphasis on “Vehicle Age.” The results highlight how predictive models reveal complex patterns beyond human intuition, improving transparency and decision-making. The proposed approach is replicable and contributes to risk management optimization in VPAs.

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

Multicriteria Analysis with Subjective Weights Versus Data-Driven Weights: An Explanatory Approach with SHAP for Risk Management

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Amably Platen,
  • Willard Silva Ribeiro,
  • Fabio Sartori Piran,
  • Daniel Pacheco Lacerda

摘要

This study compares subjective expert-based methods and data-driven predictive models for risk classification in Vehicle Protection Associations (VPAs). Using a dataset of 2,880 member records and fivefold cross-validation, Gradient Boosting achieved the best performance (86.59% accuracy, 82.50% recall), surpassing Logistic Regression, Artificial Neural Networks (ANNs), and the Analytic Hierarchy Process (AHP). SHAP analysis identified “Monthly Fee” and “FIPE Value” as key variables, contrasting with experts’ emphasis on “Vehicle Age.” The results highlight how predictive models reveal complex patterns beyond human intuition, improving transparency and decision-making. The proposed approach is replicable and contributes to risk management optimization in VPAs.