Outlier detection by rareness assumption
Tomas Hrycej and Jochen Hipp
A concept for identification of candidates for outliers is presented, with a focus on nominal variables. The database concerned is searched for rules that are almost universally valid, with rare exceptions. In statistical terms, for these rules, the hypothesis that the rule is universally valid except for random faults cannot be rejected. Outlier candidates are those values that violate these rules.
Full Text: PDF