Frequent item set-based outlier detection targets primarily categorical data by finding common values or value combinations, then identifying rows containing unusually few or infrequent patterns. section 3.2.4
Frequent pattern outlier factor (FPOF) is the simpler major approach: inliers contain multiple frequent item sets, whereas outliers contain fewer and less frequent sets. Numeric and date fields require categorical binning first. section 3.2.4
A practical workflow includes:
- Convert numeric and date values into interpretable bins. section 3.2.4
- Score each row from the frequent item sets it contains, treating rows with unusually weak pattern support as candidate outliers. section 3.2.4
Expense and SpeedDating data
Expense records reveal frequent combinations such as a staff ID appearing with an account. A SpeedDating example mines preference bins at 0.3 minimum support; the strongest inlier and outlier rows receive contrasting FPOF scores. section 3.2.4
