Scholarlink

The task of predicting the occurrence of insurance claims is important for risk management in the insurance industry. This paper considers the optimization of a decision tree model in order to improve prediction. The research applies four tuning methods: GridSearchCV, RandomizedSearchCV, HalvingGridSearchCV, and Optuna, using a publicly available dataset of insurance records. The dataset underwent preprocessing, including feature scaling and data splitting, followed by model training and evaluation. The performance of the optimized Decision Tree model was analyzed and compared based on accuracy and execution time. The results showed that HalvingGridSearchCV achieved the highest accuracy 0.983, though at the cost of increased computational time 3.90 s. RandomizedSearchCV was the fastest method, 0.20 s, while maintaining a high accuracy of 0.97, making it suitable for large datasets. GridSearchCV provided comparable accuracy 0.975 but required more time than RandomizedSearchCV. Optuna, which implements a heuristic hyperparameter optimization method based on the Tree-structured Parzen Estimator, achieved test accuracy 0.976 with an execution time 1.15 s, making it a viable alternative for complex hyperparameter spaces. The study highlights the trade-offs between accuracy and computational efficiency in hyperparameter optimization. The findings confirm that HalvingGridSearchCV is preferable when accuracy is the top priority, whereas RandomizedSearchCV is ideal for time-sensitive tasks. Optuna and GridSearchCV offer a balance between precision and efficiency. For comparison, a neural network based on the PyTorch framework was implemented and achieved an accuracy of 0.95. Slightly lagging behind the decision tree model in terms of accuracy, the PyTorch-based neural network is well suited for solving the tasks of predicting insurance claims. It should be noted that there are limitations regarding input data and interpretability.

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

Prediction of Insurance Claim Risk Using an Optimized Decision Tree Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Oleg Pursky,
  • Tetyana Filimonova,
  • Darina Bondarenko,
  • Tetyana Tomashevska,
  • Pavlo Demidov,
  • Ihor Tyschenko

摘要

The task of predicting the occurrence of insurance claims is important for risk management in the insurance industry. This paper considers the optimization of a decision tree model in order to improve prediction. The research applies four tuning methods: GridSearchCV, RandomizedSearchCV, HalvingGridSearchCV, and Optuna, using a publicly available dataset of insurance records. The dataset underwent preprocessing, including feature scaling and data splitting, followed by model training and evaluation. The performance of the optimized Decision Tree model was analyzed and compared based on accuracy and execution time. The results showed that HalvingGridSearchCV achieved the highest accuracy 0.983, though at the cost of increased computational time 3.90 s. RandomizedSearchCV was the fastest method, 0.20 s, while maintaining a high accuracy of 0.97, making it suitable for large datasets. GridSearchCV provided comparable accuracy 0.975 but required more time than RandomizedSearchCV. Optuna, which implements a heuristic hyperparameter optimization method based on the Tree-structured Parzen Estimator, achieved test accuracy 0.976 with an execution time 1.15 s, making it a viable alternative for complex hyperparameter spaces. The study highlights the trade-offs between accuracy and computational efficiency in hyperparameter optimization. The findings confirm that HalvingGridSearchCV is preferable when accuracy is the top priority, whereas RandomizedSearchCV is ideal for time-sensitive tasks. Optuna and GridSearchCV offer a balance between precision and efficiency. For comparison, a neural network based on the PyTorch framework was implemented and achieved an accuracy of 0.95. Slightly lagging behind the decision tree model in terms of accuracy, the PyTorch-based neural network is well suited for solving the tasks of predicting insurance claims. It should be noted that there are limitations regarding input data and interpretability.