Efficiently mining maximal frequent sets for discovering association rules

We present Metamorphosis, an algorithm for mining maximal frequent sets (MFS) using data transformations. Metamorphosis efficiently transforms the dataset to maximum collapsible and compressible (MC2) format and employs a top down strategy with phased bottom up search for mining MFS. Using the chess and connect dataset [benchmark datasets created by Univ. of California, Irvine], we demonstrate that our algorithm offers better performance in mining MFS compared to dGenMax (an algorithm that offers better performance compared to other known algorithms) at higher support levels. Furthermore, we evaluate our algorithm for mining Top-K maximal frequent sets in chess and connect datasets. Our algorithm is especially efficient when the maximal frequent sets are longer.