Real-time streaming mobility analytics
In our research we tested distributed streaming algorithms
and infrastructures to process large scale mobility data for fast
reaction time prediction. We use the D4D Challenge data set as a source
to generate, by multiplying with noise, even larger realistic data
sets.
Instead of addressing the problem of identifying the exact location
and movement of an individual user (that the data set is not sufficiently
detailed for), we learn global patterns both on the user level (home,
work location, daily routes) and the traffic (typical routes at time
of the day).
As a key performance indicator of our applications, we measure the running time and
the error of predictions in short range (5 minutes to 1 hour) and long
range (daily, weekly) of the location of an individual user and the
density in a given area. Over a cluster of a few old dual core servers, we are
capable of processing tens of thousands of record in a second. Our results open
the possibility for efficient real time mobility predictions of even
large metropolitan areas as well.
We demonstrate our solution via a fast reaction visual dashboard application
that can form the base of emergency or rescue services as well as
provide grounds for ride sharing, traffic planning and optimization,
thus saving natural resources.