Convolutional Neural Network Case Studies: (1) Anomalies in Mortality Rates (2) Image Recognition

Abstract

We provide a general introduction to convolutional neural networks (CNNs) in this tutorial. CNNs are particularly well suited to find common spatial structure in images or time series. As an insurance related example for life & health insurance we illustrate how to use a CNN to detect anomalies in mortality rates taken from the Human Mortality Database (HMD); the anomalies are caused by migration between countries and other errors. As a second example, we study a CNN to classify images of handwritten digits taken from one of the most widely used benchmark datasets, the Modified National Institute of Standards and Technology (MNIST) dataset. Our aim is to explore and discuss the building blocks and the properties of these CNNs, and we showcase their use.

Keywords: convolutional neural network, CNN, regression, classification, mortality rates

JEL Classification: G22