At the highest level, our team is determined to translate wireless network data into something useable, to turn computational pings into a readable narrative. Instead of attempting to map human migration patterns explicitly and then deriving some insight from the forest of routes, it is this project’s imperative to consider the metrics that can inform real-world action without following anybody’s footsteps. Technically speaking,
This platform turns raw WiFi data into high-level metrics. These high-level metrics are used to group locations and events by similarity. Using an interface to view these groups, an analyst can reparameterize the grouping criteria to explore spatial and temporal trends.
Importantly, it is not enough to visualize wireless usage or mobility patterns verbatim (er.. verdatum). Instead, we are interested in enabling the users of our platform to explore specific problems of their choice. For instance, if you would like to justify a special bus route between Tech Square and the Student Commons, you can select these locations and investigate their wireless mobility trends over time. Or, if you need to see when employers have come to visit the College of Computing building, you can select a known window of employer visitation to retrieve all times when WiFi behavior was significantly similar.
The idea is to focus on places and timeperiods, not high-definition readouts of people’s locations. With this mentality, our platform should be usable in certain specific ways:
1. cluster by days (done) As described in a previous blog post, we have already classified days by their 24 hour wireless patterns, and successfully retrieved some insightful clusters that point to daily events. This is useful as a broad, coarse overview of which days are interesting and distinct, and also has its place as a default visualization to present to analysts before they begin analysis.
2. focus on events Next we need to be able to focus in on particular time periods, and then find other times where WiFi data is similar (like the aforementioned employers in the College of Computing example). In a sense, instead of ambiguously classifying days by all of their 24-hour characteristics, this is classifying days by whether a more specific behavior is present. As we are cleaning and processing our datasets, we are also working to determine what kinds of data structures will make operations like this most efficient.
3. focus on locations Similarly, as in the case with the Tech Square to Student Common bus route, it is equally useful to be able to focus on particular locations, and see temporal trends about only these buildings.
4. focus on both Ultimately, combining event and location focusing will give our platform the greatest adaptability to use cases. And, as a final step, if we can manage to include demographic information in these results, the applications to policymaking are endless.
More on these latter components another time. For now, back to developing the data-to-insight pipeline.