Hi!

My previous/alt account is yetAnotherUser@feddit.de which will be abandoned soon.

  • 0 Posts
  • 89 Comments
Joined 1 year ago
cake
Cake day: June 1st, 2024

help-circle



  • The idea of anomaly detection is to project some input onto a (high dimensional), numeric output. From the training data alone, you can then see where the projections are clustered and develop a high dimensional “boundary” where everything within is known and good and everything outside is unknown and possibly bad. Since orders come in relatively slow, a human would be able to check for false positives and overwrite the computer decision.

    By the way, an ideal training set is preprocessed and has duplicates removed and new orders added by recombining parts of individual orders.

    For example, if we have 3 orders:

    • (Hamburger, Fries)
    • (Hamburger, Fries)
    • (Cheeseburger, Sandwich)

    We could then create the following set:

    • (Hamburger)
    • (Cheeseburger)
    • (Fries)
    • (Sandwich)
    • (Hamburger, Fries)
    • (Hamburger, Cheeseburger)
    • (Hamburger, Sandwich)

    And so on, and so forth. A naive variant is just taking the power set of all valid orders.


  • There are machine learning algorithms for anomaly detection though. They actually work decently well because exploits like this do in fact differ significantly from regular orders. Because they assume all anomalies are attempted exploits, their false negative rate is rather low while their false positive rate can be a bit higher.

    Taco Bell has the capability to create a decently large training set from all recorded orders (which must all be valid and non-malicious) so they shouldn’t have too many issues developing this model.

    If an anomaly is detected, make a human verify it is indeed an irregular order.