MULTI-DIMENSIONAL MODELING OF FUEL AND ENVIRONMENT CHARACTERISTICS

The disclosed embodiments provide techniques for generating a fuel score for a driver or vehicle. In one embodiment, a method is disclosed comprising generating a deviation vector based on recorded vehicle metrics for a given user, wherein the deviation vector comprises differences between recorded non-fuel metrics and expected non-fuel metrics; predicting, using a statistical learning (e.g., linear regression) model, deviation in fuel consumption amount using the deviation vector of non-fuel metrics; and computing a fuel score for the given user based on the deviation in fuel consumption and a current fuel score.