Mining rules to predict anomalies in the field of insurance industry from unstructured data based on data mining

Abstract

In order to help insurance industry to predict anomalies in new transactions through mining rules from a great deal of raw unstructured data, a complete, effective rule mining system is needed. In this paper, unstructured data is processed into feature vectors, which are then clustered. Clusters are used to construct a tree classifier, which contributes to reprocessing the feature vector and extracting anomaly rules. Besides, considering the weakness of a single process system, we adopt an iteration idea, in other words, we iterate the above steps for several times, thus guaranteeing the quality of the rules mined. Differently, we only focus on the information we need of much unstructured data, which avoid dealing with the whole unstructured data. Besides, combined with data mining, algorithm to extract unstructured data can achieve better effect. © 2012 Springer-Verlag.