Recently, there has been a lot of attention given to autonomous vehicles and intelligent vehicular systems. Fully integrating intelligent systems into modern vehicles relies heavily on the ability to understand and model driver behavior. Driver models are usually created using simulated driving conditions and external sensing technology. This work attempts to create an effective model for describing driving behavior using only Controller Area Network Bus (CAN-BUS) signals from a fleet of vehicles. The model consists of three layers, each meant to describe a driven trip at a different granularity. The foundational layer determines the fundamental driving maneuver, or driveme, that is being performed at any given point in the trip. The middle layer considers those drivemes to predict the driving contexts experienced on the trip. The top layer describes the trip as a whole with a simple statement about its predominant purpose. Our findings show that patterns in each of these layers can be learned and reliably predicted using a combination of decision forests and hidden Markov models, using vehicle speed and steering wheel angle input signals. We also show that potentially incorrect label predictions can be discovered using anomalous state detection. Our model is able to comprehensively describe a trip with predicted labels, providing beneficial insight into driver behavior using real-world data that is easily obtained from any vehicle on the road today.