Database Reference
In-Depth Information
Variant 2: Hiding sensitive association rules
We assume that we are provided with a database D
O
, consisting of N transactions,
and thresholds mfreq and mconf set by the owner of the data. After performing
association rule mining in D
O
using thresholds mfreq and mconf, we yield a set
of association rules, denoted as R, among which a subset R
S
of R contains rules
which are considered to be sensitive from the owner's perspective.
Given the set of sensitive association rulesR
S
, the goal of association rule hiding
methodologies is to construct a new, sanitized database D from D
O
, which achieves
to protect the sensitive association rules R
S
from disclosure, while minimally af-
fecting the nonsensitive rules existing in R (i.e., those in RR
S
). The hiding of a
sensitive association rule corresponds to a lowering of its significance, depicted in
terms of support or confidence, in the resulting database. To hide a sensitive rule, the
privacy preserving algorithm modifies the original database D
O
in such a way that
when the sanitized databaseD is mined at the same (or a higher) levels of confidence
and support, the association rules that are discovered are all nonsensitive.