Using SOM for Mortality Projection

Abstract

In this chapter, we discuss a framework based on self-organizing maps (SOMs) to explore the behavior of populations mortality rates and life expectancy. In particular, we show how to employ SOM clustering capabilities to construct coherent mortality rates, i.e., mortality rates that can be applied unchanged to a wide range of countries. In this way, we highlight that a data mining approach can be meaningful to build mortality forecasts. Besides our method is less pretending than traditional techniques in terms of both computing time and parameters to estimate. This issue is very important, provided that mortality forecasts are widely employed to develop insurance products. On the other hand, we will show that SOM clustering can be very effective to extract similar mortality patterns from apparently very different countries, thus highlighting nonlinear hidden features that are missing for more standard techniques.