Try It Yourself

Try It Yourself

animation
location
privacy
NoteTry It Yourself

Hands-on exercises to understand location privacy:

Exercise 1: Test Re-Identification Risk

Create a simple “anonymous” location dataset:

Record your GPS location every 30 minutes for 3 days (remove your name/ID). Plot locations on map with timestamps. Ask friend to identify: Where do you live? Where do you work? Where do you shop? Calculate: How many unique location-time pairs needed for identification?

What to observe: Most people can be uniquely identified from just 4-5 location points (MIT study). Morning residential cluster = home. Daytime cluster = work. Evening+weekend patterns reveal hobbies, relationships, health visits. Even “anonymous” data reveals identity.

Exercise 2: Implement Tiered Disclosure

Build a geofence system with privacy tiers:

  1. Tier 0: No location shared (default)
  2. Tier 1: Boolean home/away only (on geofence transition)
  3. Tier 2: Zone name (“Downtown”, “Home neighborhood”)
  4. Tier 3: Coarse location (100m radius circle)
  5. Tier 4: Precise coordinates (emergency only)

Test each tier: What minimum data satisfies functional need (e.g., “preheat home”)?

What to observe: Most automation needs Tier 1-2 only. Tier 4 precise coordinates rarely necessary outside emergencies. Users accept Tier 1-2 but distrust Tier 4 continuous tracking. Design principle: collect minimum precision needed.

Exercise 3: Calculate K-Anonymity

Analyze a location dataset for anonymity:

Take sample dataset: [{time: “9am Monday”, location: “Coffee Shop A”}, …]. For each unique location-time pair, count how many people visited. If count < k (e.g., k=5), suppress that data point. Analyze: Under your chosen scenario assumptions, what percentage of records would be suppressed for k=5?

What to observe: In this sample scenario, rare locations such as a specialist doctor, support group, or late-night visit may have low counts and be suppressed. The chosen percentage is an input assumption; actual loss and re-identification risk depend on the dataset and release design.

Back to the chapter