One scan based high average-utility pattern mining in static and dynamic databases

摘要

High average utility pattern mining has been proposed to overcome the demerits of high utility pattern mining. Since high average utility pattern mining can extract more valuable patterns than high utility pattern mining, many related researches are being actively conducted. However, most studies in high average utility pattern mining have only focused on mining in static databases not dynamic databases. In addition, the methods of previous studies with dynamic databases consume a huge runtime and memory space due to the inefficient processes and structures. To overcome these problems, we present a novel high average utility pattern mining approach from the dynamic databases. The proposed mining approach reads a database only once and adopts a new data structure called a HAUP-List to store information of patterns more compactly. In addition, in order to reflect the incremental environments, a restructure process is designed to handle the newly inserted data. Thus, our approach can extract high average utility patterns more efficiently than the suggested methods in previous works in dynamic databases. Various experiments are conducted to demonstrate the performance of the proposed approach using both real and synthetic datasets. Results of these experiments show that the proposed mining approach outperforms the other state-of-the-art high average utility pattern mining approaches in dynamic databases.