Device Recurring Areas and Sensitive Locations¶
Defining recurring areas¶
We define a recurring area based on:
- The number of days spent at a given location in the last month
- The daily average number of hours spent at that location
- The time of the day spent at the location (nighttime/daytime)
The name “recurring area” is self-explaining, but to give a bit more detail, recurring areas for a device may be its home and work locations, given that a device’s owner can work from home and we cannot say why that device is staying in the area our algorithm has identified as its home.
The algorithm's first goal is to identify the home area of a device. A score is computed for each location point of that device, by combining the three variables listed above. This score represents the probability that the considered location point is within that device's home area. The more days and the higher the average of hours the device spent in that area, especially during nighttime, the higher the score. The area with the highest score is selected as the device's home area.
Once the home area is identified, the algorithm runs to assign a work area to the device. Note that the algorithm requires the work area to be located at least 100 meters apart from the home area. The algorithm uses the same variables used for detecting the home area, but it assigns a higher score to locations where the device dwells more during the daytime.
The recurring area algorithm runs daily, to confirm or update the inferred recurring area as we observe new data.
The algorithm only considers devices present in the Spectus user base for at least 2 days.
How we expose the device's recurring areas¶
In order to preserve the users' privacy, Spectus tightly restricts access to the inferred precise locations within the recurring areas (i.e. the home and work locations) of devices. Furthermore, inferred recurring areas are used as input on various downstream processes, in order to create more privacy-protected versions of the Spectus datasets.
For example, we only expose the recurring areas at the Census Block Groups level, rather than the precise locations, within our Platform. This allows for reaching a good balance between utility and privacy. Indeed, according to the U.S. Census Bureau, there are between 600 and 3,000 people living in each block group. Each block group is an aggregate of contiguous U.S. blocks sharing similar socio-demographic characteristics.
For other countries, we expose the recurring areas at the level 6 geohash, which provides an obfuscation error around the actual location of about ±0.31 Kilometers.
In addition to exposing only a user's inferred recurring area Census Block Group and not their precise home or work location, all data points and stops in the vicinity of the user inferred recurring area are transformed to the centroid of this Census Block Group. This enhances the privacy of the device and reduces the risk of individual identification (have a look at our Privacy Enhanced Mobility white paper) This process is referred to as "upleveling". When a user is close to his inferred recurring area, the system classifies the corresponding mobility points as "RECURRING_AREA". All RECURRING_AREA points are UPLEVELED and are labeled as such in the transformation_type column of the corresponding table.
From August 30th, 2022 an adaptive algorithm has replaced the static upleveling zone around a device's home or work location. This new algorithm varies the shape and size of the geographical area to be upleveled by taking into account population density.
Sensitive locations¶
In addition to the devices’ recurring area, we remove from all our datasets any mobility point falling close enough to a sensitive POI.
Examples of categories of sensitive POIs are:
- Healthcare Related Sites
- Child Related Facilities
- Churches/Religious Facilities
- Federal/Government Buildings
- Military Related Facilities
- Payday Loan Services
- Ethnic Minorities Related Facilities
For more details, please have a look at our Sensitive POIs Policy on the Spectus website.