An Efficient Bit-Based Approach for Mining Skyline Periodic Itemset Patterns

Abstract

Periodic itemset patterns (PIPs) are widely used in predicting the occurrence of periodic events. However, extensive redundancy arises due to a large number of patterns. Mining skyline periodic itemset patterns (SPIPs) can reduce the number of PIPs and guarantee the accuracy of prediction. The existing SPIP mining algorithm uses FP-Growth to generate frequent patterns (FPs), and then identify SPIPs from FPs. Such separate steps lead to a massive time consumption, so we propose an efficient bit-based approach named BitSPIM to mine SPIPs. The proposed method introduces efficient bitwise representations and makes full use of the data obtained in the previous steps to accelerate the identification of SPIPs. A novel cutting mechanism is applied to eliminate unnecessary steps. A series of comparative experiments were conducted on various datasets with different attributes to verify the efficiency of BitSPIM. The experiment results demonstrate that our algorithm significantly outperforms the latest SPIP mining approach.