Assessing driving risk through unsupervised detection of anomalies in telematics time series data (Q6959423)

The authors of the paper proposed a \textit{continuous-time hidden Markov model} framework for modelling trip-level vehicle telematics data, enabling analysis and driving risk assessment at both trip and driver levels. Through a detailed discussion of the telematics data structure and challenges such as irregular time intervals and varying trip lengths, the authors have highlighted the complexities involved in processing and modelling telematics data. They stated and explained the key adjustments to the \textit{continuous-time hidden Markov model} and necessary preparation of the telematics data to ensure effective modelling and analysis.\N\NHaving chosen a rather complex probabilistic model, the authors solve three specific problems in their work. First, they define an anomaly index to quantify the level of anomalousness of each trip, capturing deviations from normal driving behavior. Second, by detecting anomalous driving behavior, they aim to identify trips associated with accidents. Third, the authors evaluate risk at both the trip and driver levels because not all trips by a claimed driver are dangerous, nor are all trips by a no-claim driver safe.