Cost-sensitive Granular Computing-based Identification Method for Insurance Fraud Data

Given the large scale and uneven distribution of insurance claim prediction data, traditional methods incur high information costs in identifying fraud samples. To improve fraud detection performance, this study proposes a new artificial intelligence approach based on Granular Computing theory and cost sensitive learning. First, the data is finely granulated based on categorical features and purity levels. Second, a multi-granularity information representation system is created to capture the data’ s underlying structures and features. Finally, a cost matrix mechanism is introduced to optimize the classifier, improving both the accuracy and effciency of fraud detection. Experimental results show that the proposed method significantly improves detection precision. It provides strong technical support for effectively curbing fraud, protecting the legal rights of stakeholders, and maintaining market order.