Health-policyholder clustering using health consumption

On paper, prevention appears to be a good complement to health insurance. However, its implementation is often costly. To maximize the impact and efficiency of prevention plans these should target particular groups of policyholders. In this article, we propose a way of clustering policyholders that could be a starting point for the targeting of prevention plans. This two-step method mainly classifies using policyholder health consumption. This dimension is first reduced using a Nonnegative matrix factorization algorithm, producing intermediate health-product clusters. We then cluster using Kohonen's map algorithm. This leads to a natural visualization of the results, allowing the simple comparison of results from different databases. We apply our method to two real health-insurer datasets. We carry out a number of tests (including tests on a text-mining database) of method stability and clustering ability. The method is shown to be stable, easily-understandable, and able to cluster most policyholders efficiently.