In various embodiments, the output of a predictive model, for example, of any of the 2D, 3D, or ensemble prediction models as described herein, can provide or be used to generate a data structure or series of data structures relating to or descriptive of a driver's behavior. For example, FIG. 10 illustrates a text-based data structure 1000 that may be output from a predictive model based on a series of frames, images or chunks input into the model, where the series of frames or images relate to a driver operating a vehicle. The text-based output data structure 1000 can be, for example, in the Javascript Object Notation (JSON) format or some other text-based format. In other embodiments, the output format can be binary, for example, a serialized “struct” object of the C++ or C# programming language. The structure can also be non-text-based, for example, based on an numeric enumeration, where each number in the enumeration can indicate a specific value, e.g., the enumeration value “1” for “Normal Driving” and the enumeration value “2” for “Texting.”
