Traditional time-of-arrival estimation techniques are often used in advanced transportation systems such as in-car navigation systems and fleet management systems. These techniques tend to neglect certain categories of context when estimating a vehicle's time-of-arrival at a given destination. Particularly important categories of such neglected context include driver behavior patterns and traffic flow patterns. Failure to consider such context can significantly reduce the accuracy of time-of-arrival estimations. This paper presents a technique that considerably increases accuracy in time-of-arrival estimations. Our technique achieves this increase in accuracy by taking into account observed patterns in traffic flow and in driver travel behavior. Additionally, this technique does not rely on map data or a connection to a remote server, but can run autonomously and with a minimum amount of initialization or user input. As a result, a transportation system using our technique can provide a higher level of accuracy, pervasiveness, and privacy compared to systems based on traditional time-of-arrival techniques.