Scholarlink

The insurance industry’s capacity to forecast future claims pertaining to specific policyholders is one of its major problems. The industry has struggled to use several risk indicators to correctly anticipate the likelihood of policyholder claims using historical data since risk varies from policyholder to policyholder. This study used a deep learning model of ResNet with activation functions Softmax and Leaky ReLU. The dataset consisted of just 1338 samples in total, and 17% of the samples were removed during preprocessing. Due to the extreme imbalance of the samples, the SMOTE Technique was used to balance the classes in the data, which resulted in the addition of 5.5% additional samples. R2 score, RMSE, and MPE are the performance measures used during training in order to evaluate the model’s performance. Leaky ReLU yielded values of 0.89%, 0.59%, and 0.35%, respectively, whereas the Softmax function yielded values of 0.50%, 1.10%, and 1.20%.

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

Deep Learning-Based Insurance Claim Prediction Using SMOTE and ResNet Algorithm

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ravi Kumar Ravi

摘要

The insurance industry’s capacity to forecast future claims pertaining to specific policyholders is one of its major problems. The industry has struggled to use several risk indicators to correctly anticipate the likelihood of policyholder claims using historical data since risk varies from policyholder to policyholder. This study used a deep learning model of ResNet with activation functions Softmax and Leaky ReLU. The dataset consisted of just 1338 samples in total, and 17% of the samples were removed during preprocessing. Due to the extreme imbalance of the samples, the SMOTE Technique was used to balance the classes in the data, which resulted in the addition of 5.5% additional samples. R2 score, RMSE, and MPE are the performance measures used during training in order to evaluate the model’s performance. Leaky ReLU yielded values of 0.89%, 0.59%, and 0.35%, respectively, whereas the Softmax function yielded values of 0.50%, 1.10%, and 1.20%.