The more experienced we become with traversing wifi usage data, the easier it is to expose the extremely specific user behaviors that have an uncanny resemblance to observing someone in reality. Even though the resolution of data that is readily accessible by us might be very high, the micro nature of information is not necessarily vital for us to understand high-level mobility patterns or to solve specific resource allocation policies. It is important to balance usefulness with privacy, and to find a granularity that speaks volumes of information without disclosing the private data points of its subjects.
To this end, we have defined a series of metrics that work to provide us with useful interpretations of human movement without sacrificing privacy. Although we will wait to present the finalized list of measurements when we have determined their effectiveness, our overarching goal is to determine location-based variability between days that describes people holistically, as opposed to individually. This ‘symptomatic’ approach entails that we do not need to know the lowest level of mobility detail in order to make informed decisions. Whereas a physician does not need to know the activities of every cell in the body to diagnose a patient, we do not need to the migration of every internet user to know when it is Monday, raining, or a school holiday, and even whether a route between buildings is being significantly used.
One way to create these high-level trends is be to group different days of the month based on the wireless usage pattern each of over the days. Are all Mondays similar? Is the usage of a building in the weekends similar to that during holidays? What other interesting similarities can we infer? Our visualization prototype shows usage patterns in CULC for the month of January 2014. With four groups clustered simply by K Means, we can already see that
- The second week is the busiest irrespective of the day of the week
- T/TH is distinguishable from M/W/F, in coherence with the class schedule (usage dips down earlier during the day on tuesdays and thursdays despite a higher peak)
- Before classes start, the weekends, and Martin Luther King Day ( 20 ) are the same
- The effects of Snowpocalypse are evident ( 29 / 30 )
The approach is to package these daily comparisons in such a way that campus policymakers can both inform and support their decisions with hard evidence, as well compare new unpredicted events to situations from the past.