AN EFFICIENT GRADUAL PRUNING TECHNIQUE FOR UTILITY MINING

Utility mining in knowledge discovery has recently become a prominent research issue due to its many practical applications. A high utility itemset in utility mining considers not only quantities but also profits of items in transactions. Most of previous approaches were based on the traditional utility upper bound model to find high utility itemsets in databases. By using the model, however, a huge number of candidates have to be generated, and a good deal of time to count utility upper bounds of itemsets has to be needed for mining. In this paper, we thus propose a level-wise mining approach to find efficiently high utility itemsets in databases. In particular, a pruning strategy is designed to gradually cause better utility upper bounds of itemsets in passes. Also, data size could be gradually reduced to save data scan time. Finally, the experimental results on synthetic datasets and a real dataset show the proposed approach outperforms the traditional two-phase utility mining approach in pruning effect and execution efficiency.