Characterizing Car Trips Through Information Theory Metrics

In this work, we apply information theory metrics to car trips logged by volunteers around the world and use quantifiers such as location entropy to reveal aspects of users' mobility, like the context in which trips happened. The dataset used in this work was collected from the enviroCar project and contains not only location logs but also sensor readings associated with each location. Information theory measurements can also reveal relationships between sensor measurements in order to reveal rare occurrences and reduce uncertainty. This work shows that it is possible to differentiate driving contexts and capture relationships among sensors using location entropy and mutual information, respectively. These contributions pave the way for developing new features that may ultimately improve traffic context classification results.