Combining accelerometer data and contextual variables to evaluate the risk of driver behaviour

Telemetry devices are generating and transferring increasingly more data, with notable potential for decision makers. In this paper we consider the accelerometer and speed data produced by in-vehicle data recorders as a proxy for driver behaviour. Instead of extracting harsh events to cope with the large volumes of data, we discretise the data into a tractable and finite risk space. This novel methodology allows us to track both acceptable and non-acceptable driving behaviour, and calculate a more comprehensive risk model using the envelope of the data, and nota priorithresholds. We show how thresholds suggested in literature can characterise some driving behaviour as good, even though our empirical evidence has not even registered such extreme driving behaviour. We demonstrate the model using accelerometer data from 124 vehicles over a one month period. Three rules, each a combination of accelerometer and/or speed data, are applied to the risk space to derive person-specific scores that are comparable among the individuals. The results show that the scoring is useful to identify specific risk groups. The proposed model is also dynamic in that it dynamically adjusts to the observed records, instead of data having to abide by a limited model specification.