Selectivity estimation using homogeneity measurement

A new approach is presented for organizing a large collection of multidimensional data with an unknown distribution by partitioning the data such that the data are relatively homogeneously distributed in each block. A multidimensional tree is generated according to this partition. After the tree is generated, summary data estimation such as selectively estimation can be performed via a tree search. This approach is applicable to both ordered and categorial attributes. The merits of this method are verified theoretically and by simulation.<
>