Frequent pattern mining can be a lot of patterns, but judging whether a pattern is interesting requires a pattern evaluation method. The common pattern evaluation methods are described below. (Hypothetical set of items A, B)
1. Support Degree
The ratio of the number of tuples in the item set A and B to the number of all tuples, typically P (a∪b).
2. Reliability
The confidence level of mode a--> B is P (b| A
3. Lifting Degree
Lift (A, B) = P (a∪b)/(P () *p (B)), the elevation greater than 1 is positive correlation, less than 1 is negative correlation, equals 1 is independent.
4, the card side measurement
5. Full confidence degree
All_conf (A, B) = Min{p (a| B), P (b| A)}
6. Maximum confidence level
Max_conf (A, B) = Max{p (a| B), P (b| A)}
7, Kulczynski measurement
KULC (A, B) = 1/(2* P (a| B) + P (b| A
8, cosine measurement
cosine = sqrt (P (a| B) * P (b| A))
In these measurements, only the elevation and the cardholder measures are not 0 invariant measures, but other metrics also face an imbalance problem, introducing an imbalance ratio:
Where the SUP represents the confidence level.