Sparse predictive modeling techniques with applications in insurance pricing and mortality forecasting

The insurance sector relies heavily on data for a variety of their operational processes such as product pricing, marketing and estimating future expenses. As today's society generates data more rapidly than ever before, the demand for new algorithms, able to infer meaningful information from this data, is rising. A modern issue in insurance is that data sets not only contain a lot of observations, but also many variables of different types. Many existing algorithms are only developed for specific variable types or data settings which are often not applicable to insurance practice.

The main ambition of this research is to merge and extend the existing statistics and machine learning literature to develop new algorithms leading to accurate and interpretable predictive models while taking into account the specific data issues in insurance practice. These algorithms should have a sound statistical basis while also be able to deal with the increasing data availability. This work focusses on the use of regularization to develop sparse, interpretable predictive models. The methodology has concrete applications in predictive models for insurance pricing and in simultaneously projecting future mortality of multiple populations.