Anonymize Geolocation During Data Capture

With all the news around privacy and geolocation for pandemic analysis, my mind started wandering on the general subject of geolocation randomization or rounding. The simplest, yet crude, approach is to use AppSheet’s Here() function and then lop off some decimal places. There is probably a more sophisticated way to do this (rounding on mercatur calc, etc). However, even converting a full Lat/Long to a rounded one provides some level of obscurity of geolocation, while still allowing for meaningful analysis to occur later.

Slight brag about the platform: I’m not aware of many tools or software out there that can allow for geolocation and rounding/obfuscation to occur simultaneously in realtime during data entry or data capture. Let us know if you know of some!

Anyway, here’s a quick example of what the AppSheet expression looks like:

https://www.appsheet.com/samples/Simple-calcs-to-anonymize-your-geolocation-during-data-entry?appGuidString=1dc7aef1-0a17-4256-8d39-800fbb5be361

Ty,

So I have an idea, but I don’t think appsheet is the right application for it. But it goes carries on to the idea of contact tracing. If an app had the ability to track historic location, and compile social media data (friends list), it should be able to do two things.

  1. Identify people that have indirect contact with COVID-19 by walking into the same place with a positive result.
  2. Identify people that have had likely direct contact with COVID-19. The thought is that if you and I were friends on Facebook, and at Lowes at the same time, we were probably together.

These two items could then be used to build an exposure score. # of Direct and # of In Direct exposures. If we had the ability to overlay positive cases, then should basic machine learning be able to give a prediction score on having contracted COVID-19?

I know there is a privacy concern, and you are more familiar with it then I am, but what if unique keys were assigned to each user in the application. The table we see only has the unique keys, and the names are blind to everyone.

We’d need positive COVID-19 cases to train the model, and to continue training the model. When that is built,it could then be used to identify early on if someone is infected with COVID-19 prior to them showing symptoms. This is the big part of it, if we could let people know they are at a high likelyhood for carrying, prior to being symptomatic, potentially we could assist in preventing the spread with the two weeks of not being symptomatic.

There is a HIPPA concern, but I think we get around that if people opt in to the system. The people we specifically need to opt-in are those people that were tested positive.