Parallel algorithm for mining frequent episodes

We now present an algorithm, called PRE (parallel algorithm using database reduction technique for mining frequent episodes) by us, that is more efficient than the existing WINEPI algorithm because: (1) PRE utilizes the efficiency of parallel computing and (2) the size of database can be gradually reduced during mining. In the full paper, we explain PRE algorithm in detail; in this abstract, we just add some pertinent remarks to listing the three topics of explanation: (1) the important properties in parallel mining episodes of frequent occurrence; (2) the database reduction techniques in parallel mining frequent episodes; and (3) the iterative procedure of PRE algorithm; in topic 1, we give Theorems 1, 2, and 3 in the full paper that make clear the relations among global frequent episodes and local frequent episodes under various conditions; in topic 2, we give Theorems 4 and 5 in the full paper for reducing the database gradually during mining; in topic 3, we give a four-step iterative procedure. Finally we give some numerical examples, whose results, shown in Figs.1, 2, and 3 in the full paper, show preliminarily that, by using the database reduction techniques alone, algorithm PRE is faster than WINEPI about 25%. The experiment results also show that algorithm DRE has good speedup performance.