Improving Naïve Bayes models of insurance risk by unsupervised classification

In the paper application of Naive Bayes model, for evaluation of the risk connected with life insurance of customers, is considered. Clients are classified into groups of different insurance risk levels. There is proposed to improve the efficiency of classification by using cluster analysis in the preprocessing phase. Experiments showed that, however the percentage of correctly qualified instances is satisfactory in case of Naive Bayes classification, but the use of cluster analysis and building separate models for different groups of clients improve significantly the accuracy of classification. Finally, there is discussed increasing of efficiency by using cluster validation techniques or tolerance threshold that enables obtaining clusters of very good quality.