A frequent item set is a value within one column, or a combination of values across columns, that appears together in a substantial fraction of records. It is useful for primarily categorical data, where numeric encoding may be inconvenient. section 3.2.4
For outlier detection, frequent pattern outlier factor (FPOF) counts the frequent item sets contained in each row and considers rows containing fewer and less frequent sets more suspicious than typical rows. Association rules provide another frequent-item-set method but are treated separately. section 3.2.4
A practical FIS-based detection workflow is:
- Convert dates and numeric values into categorical bins, choosing binning schemes that preserve interpretability. section 3.2.4
- Inspect flagged rows as candidates rather than treating rarity alone as proof of meaningful abnormality. section 3.2.4
Staff expenses and SpeedDating
In staff expenses, repeated combinations such as Staff ID 9000483 with Account Meals form a frequent pattern. In SpeedDating, item sets with minimum support 0.3 include ambitious ratings [0-15] with support 0.797326 and the combination of ambitious and shared-interests ratings with support 0.593817. The listed most-inlier and most-outlier rows have FPOF scores 0.0 and 1.0, respectively.
