February 2018
Intermediate to advanced
378 pages
10h 14m
English
Conviction is a measure that helps to judge if the rule happened to be there by chance or not. It was introduced by Sergey Brin and coauthors in 1997 [1] as a replacement for confidence, which can't capture the direction of an association. Conviction is a comparison of the probability of if appearing without then, if they were dependent on the actual frequency of if without then:
![]()
In the nominator, we have the expected frequency of item sets without {
}. (In other words, how often the rule doesn't hold true.) In ...
Read now
Unlock full access