Learning Rare Class Footprints: the REFLEX Algorithm
Ray J. Hickey, Northern Ireland · 2003
An r-contour footprint is a set of individuals each of whom has a propensity of at least r of belonging to a rare class. The properties of footprints are summarized. An algorithm, REFLEX, is proposed for extracting a footprint from an induced decision tree. Results of initial experiments comparing REFLEX to mestimation Laplace smoothing show that both algorithms deliver broadly similar performance for different contours. Unlike Laplace, REFLEX does not require extensive tuning. When high propensity rare class disjuncts exist (> 50%), both algorithms perform better on pruned trees.