Claims Frequency Modeling Using Telematics Car Driving Data

We investigate the predictive power of covariates extracted from telematics car driving data using the speed-acceleration heatmaps of Gao and Wüthrich (2017). These telematics covariates include K-means classi cation, principal components, and bottleneck activations from a bottleneck neural network. In the conducted case study it turns out that the first principal component and the bottleneck activations give a better out-of-sample prediction for claims frequencies than other traditional pricing factors such as driver's age. Based on these numerical examples we recommend the use of these telematics covariates for car insurance pricing