Data mining in insurance claims (DMICS) two-way mining for extreme values

In insurance claims extreme values are inevitable and cannot be discarded for predictive model building. Moreover, settling insurance claims involves many objections, human sentiments and unseen factors which are hard to be estimated. This simple fact presents the greatest challenge to analysts working on such problems. This paper presents an optimal approach to minimize the effects of this problem on predictive analysis. The data in question includes insurance settlement cases. The proposed approach firstly deals with extreme values by classifying them separately and then applies machine learning models for predicting settlement amount for each claim. This two way mining increases the overall accuracy in predicting settlement amount for insurance claims.