next up previous
Next: Subgroup Evaluation Measures Up: Related Work Previous: Subgroup Discovery

5.2 Measures of Interestingness

Various rule evaluation measures and heuristics have been studied for subgroup discovery, aimed at balancing the size of a group (referred to as factor g by Klosgen1996) with its distributional unusualness (referred to as factor p). The properties of functions that combine these two factors have been extensively studied (the so-called ``p-g-space'').

Similarly, the weighted relative accuracy heuristic, defined as $WRAcc(Class \leftarrow Cond) = p(Cond)\cdot (p(Class\vert Cond)-p(Class))$ and used by tod:fla:lav:2000, trades off generality of the rule (p(Cond), i.e., rule coverage) and relative accuracy
p(Class|Cond) - p(Class). This heuristic is a reformulation of one of the measures used in EXPLORA.

Besides such `objective' measures of interestingness, some `subjective' measure of interestingness of discovered patterns can be taken into the account, such as actionability (`a pattern is interesting if the user can do something with it to his or her advantage') and unexpectedness (``a pattern is interesting to the user if it is surprising to the user'') ( Silberschatz & Tuzhilin, 1995).