Developing an Abnormal Pattern Classification Model based on Secondary Abnormal Detection
The scale of insurance fraud and unfair claims is increasing every year, so emerging as a social problem. However, the unfair claim detection technology is not sufficient to detect increasing unfair claims and intelligent insurance fraud patterns. So, it is necessary to improve the technology. This study developed a secondary abnormal detection model classifying newly patterns in using the Isolation Forest algorithm and the K-means clustering algorithm sequently as a method for quickly finding out increasing unfair claims in the insurance big data and efficiently detecting unfair claim patterns. The binary labels are divided into normal and abnormal data in the existing data through the primary and secondary abnormal detections, and the both labels are changed into 4 multiple labels, specifying the characteristics of data group. The characteristics are applied to various classification algorithms to measure the classification performance of each algorithm. Also, this study identifies the detection results of unfair claims of insurance data, and tests the secondary abnormal detection models’ overdue data detection performance by applying the models to the overdue cases data for rental companies. As the results from test, it is found that the developed models can improve the unfair claim accuracy of life insurance by 92% and the overdue detection accuracy of rental fee can increase by 96%.