Tree Growth Based Episode Mining without Candidate Generation

Mining for frequent episodes has been an active research area in recent years. Numerous algorithms have been developed to discover different types of episodes, where most of them adopt an a priori-like approach that generates candidates and then recognises these candidates to determine their support. However, such methods are computationally expensive, depending on the size and structure of the input data. Within this paper a tree growth based method is presented discovering episodes without candidate generation. The presented method only consists of a recognition phase that dynamically extends a specialised tree structure to efficiently store and process episodes.