Evaluation and aggregation of pay-as-you-drive insurance rate factors: A classification analysis approach

Vehicle sensor data enable novel, usage-based insurance premium models known as 'Pay-As-You-Drive' (PAYD) insurance, but pose substantial challenges for actuarial decision-making because of their inherent complexity and volume. Based on a large real-world sample of location data from 1572 vehicles, the present study proposes a classification analysis approach that addresses (i) the selection of predictor variables, (ii) the presence of class skew and time-variant prior distributions, and (iii) the suitability of classifier scores as an aggregated actuarial rate factor. Using raw location data, we derive a set of 15 predictor variables that we use to train and compare logistic regression, neural network, and decision tree classifiers. We find that while neural networks exhibit superior classification performance, logistic regression is better suited from an actuarial viewpoint in several ways. In sum, our results clearly demonstrate the potential of high-resolution exposure data for reducing the complexity of PAYD insurance pricing in practice.