XGBoost is recognized as an algorithm with exceptional predictive capacity. Models for a binary response indicating the existence of accident claims vs. no claims can be used to identify the determinants of traffic accidents. We compare the relative performances of logistic regression and XGBoost approaches for predicting the existence of accident claims using telematics data. The dataset contains information from an insurance company about individuals’ driving patterns – including total annual distance driven and percentage of total distance driven in urban areas. Our findings show that logistic regression is a suitable model given its interpretability and good predictive capacity. XGBoost requires numerous model-tuning procedures to match the predictive performance of the logistic regression model and greater effort as regards interpretation.

We analyze real telematics information for a sample of drivers with usage-based insurance policies. We examine the statistical distribution of distance driven above the posted speed limit – which presents a strong positive asymmetry – using quantile regression models. We find that, at different percentile levels, the distance driven at speeds above the posted limit depends on total distance driven and, more generally, on such factors as the percentages of urban and nighttime driving and on the driver’s gender. However, the impact of these covariates differs according to the percentile level. We stress the importance of understanding telematics information, which should not be limited to simply characterizing average drivers, but can be useful for signaling dangerous driving by predicting quantiles associated with specific driver characteristics. We conclude that the risk of driving long distances above the speed limit is heterogeneous and, moreover, we show that prevention campaigns should target primarily male, non-urban drivers, especially if they present a high percentage of nighttime driving.