Efficient approach of recent high utility stream pattern mining with indexed list structure and pruning strategy considering arrival times of transactions

One of various pattern mining techniques, the High Utility Pattern Mining (HUPM) is a method for finding meaningful patterns from non-binary databases by considering the characteristics of the items. Recently, new data continues to flow over time in diverse fields such as sales data of market, heartbeat sensor data, and social network service. Since these data have a feature that recently generated data have higher influence than the old data, research has been focused on how to efficiently extract hidden knowledge from time-sensitive databases. In this paper, we propose indexed list-based algorithm that mines recent high utility pattern considering the arrival time of inserted data in an environment where new data is continuously accumulated. In other words, to treat the importance of recent data higher than the that of old data, our algorithms reduces the utility values of old transactions according to the time the data is inserted by applying damped window model concept. Moreover, we carry out various experiments to compare our method with state-of-the-art algorithms using real and synthetic datasets in diverse circumstances. Experimental results show that our algorithm outperforms competitors in terms of execution time, memory usage, and scalability test.