Iceberg-Cube Algorithms for Finding Frequent Combinations of Values: Experiments on Synthetic and Real Databases

L. Findlater and H.J. Hamilton (Canada)


machine learning, data mining,knowledge discovery, data cube, bottomup computation,topdown computation.


Two straightforward approaches to computing Iceberg Cubes are bottom-up computation and top-down computation. Previously, results based on a single database showed that the bottom-up method is more efficient than the top-down method, even when a pruning capability was added to the top-down method. Here we report on more extensive experiments featuring a variety of synthetic and real databases. Results show that a crossover point occurs at a low minimum support threshold, below which the top-down method is more efficient. The bottom-up method grows increasingly inefficient as the number of results increases.

Important Links:

Go Back