Frequent 2-Episode Mining with Minimal Occurrences Based on Episode Matrix and Lock State

Frequent episode mining helps to set up episode rules and predict future events. In frequent episode mining, 2-episode mining plays an important role. The mining methods for 2-episodes determine the global strategies of frequent episode mining. The paper focuses on minimal occurrence based frequent 2-episode mining. For the problems existing in the current methods, a novel frequent 2-episode mining method is proposed with high efficiency based on episode matrix and the lock strategy. It does not need to generate candidate episodes and only scans data once. A series of experiments on real data sets show the advantages of the proposed method at time and space cost.