Optimal Discretization of Quantitative Attributes for Association Rules [chapter]

Stefan Born, Lars Schmidt-Thieme
2004 Classification, Clustering, and Data Mining Applications  
Association rules for objects with quantitative attributes require the discretization of these attributes to limit the size of the search space. As each such discretization might collapse attribute levels that need to be distinguished for finding association rules, optimal discretization strategies are of interest. In 1996 Srikant and Agrawal formulated an information loss measure called measure of partial completeness and claimed that equidepth partitioning (i.e. discretization based on base
more » ... tervals of equal support) minimizes this measure. We prove that in many cases equidepth partitioning is not an optimal solution of the corresponding optimization problem. In simple cases an exact solution can be calculated, but in general optimization techniques have to be applied to find good solutions.
doi:10.1007/978-3-642-17103-1_28 fatcat:tkqg6oapifae7fjh5cwbbjb3da