ALIASED AS Local Recoding with zero suppression with a mix of various levels of generalization

2 quasi-identifiers - workclass, dob

The lattice structure is generated going from 0, 0 to 5, 5 where each successive element differs from the previous one by one, i.e only one qid can get updated at a singe time

Path to carry out experiments : [(0, 0), (0, 1), (0, 2), (1, 2), (1, 3), (1, 4), (1, 5), (2, 5), (3, 5), (4, 5), (5, 5)]

Algorithm:

• Generalization lattice is used here, and the concept of local recoding is applied here.

• The existence of k-anonymity is checked, for tuples that do not form a part of any equivalence class, they are generalised to the highest level, thus preserving all the tuples, and no loss of data.

• 2 variants were carried out, wherein case 1, higher levels of generalisation in the path were used, while in case 2, after every successive iteration, the path started from the lowest level of generalization

  Results for Case 1 :

  (0, 2) 4

  (1, 2) 3928

  (1, 3) 151

  (1, 5) 72

  (2, 5) 820

  (3, 5) 14

  (5, 5) 11

  Results for Case 2 :

  (0, 2) 4

  (1, 2) 3296

  (1, 3) 440

  (1, 5) 196

  (2, 5) 936

  (3, 5) 104

  (5, 5) 24

  • The results obtained on the 2 variations were not very different, but the case 1 with lesser compute can be preferred
  • No restriction on the size of k is used here, i.e an equivalence class will mandatorily have at least k-1 elements present
  • Wrt to temporal attack, cannot guarantee anything, but due to odd class sizes it is possible that there is a linking attack