Data Privacy in Practice: Pseudonymisation, k-Anonymity & Re-identification Risk

What you'll build

What you will be able to do

Each lab ends with a concrete result that you build yourself and the checker confirms, cell by cell.

• Sort the columns of the (fictional) Larkfield Community Health Survey into direct identifiers, quasi-identifiers and sensitive attributes, the first step of every anonymisation project.
• Replace the direct identifiers of ten customers with a pseudonym from a key table, build an illustrative salted ID, mask the email address and see why the key holder can still re-identify everyone.
• Coarsen two quasi-identifiers of 20 survey records, age into 10-year bands and ZIP into a 3-digit area, and count how many records are still unique.
• Measure how many records share each combination of quasi-identifiers, find the k of the release and check it against a target of k = 3.
• Suppress the records in classes below k = 3, measure how much of the data that costs and check whether it biases the statistic the researchers care about.
• Count the diagnoses inside each equivalence class of a 3-anonymous release, find the classes with only one diagnosis and measure the l of the release.
• Turn the class sizes of the release into re-identification risks, report the maximum and the average risk and chart the risk by class.
• Apply a retention schedule to twelve records: look up the retention period, calculate each delete-by date with EDATE and flag the rows that are due on the review date.