One of the challenges I encountered through the course of Intel’s LEAM project was translating from a concept as we understood it into an algorithm the computer would understand. (This is, I would say, the primary challenge of machine learning and data mining). One of those translation tasks had to do describing, mathematically, how “familiar” a participant was with the area through which they were driving.
I experimented with a variety of models, but finally settled on one that defined a participant’s familiarity with a point using two components: how close the point is to most of the points in their history, and whether they have ever visited that point before. Using that model, I visualized participant’s data as heat maps.
The familiarity model by itself is not revelatory (although it’s nice to be able to describe it mathematically!), but could be a very useful feature to use in other algorithms and analyses. Using this model, for instance, I was able to look at the relationship between how familiar a participant was with a particular area and their phone use.