A Mining Method Based On Utility Pattern From Data Stream

With the gradual establishment of data-oriented society, the role of database informationsystem in all aspects of life become important more and more. The database informationsystem maintains a large amount of data, while a lot of valuable informations difficult to findare hiding in the accumulated date. Guarantying the accuracy of this informations,how to digout them has become an important issue. While the static database can not meet the real-worldapplications, the data stream mining technology so emerged. The difference between them isthe continuity of data streams, unlimited, high-speed and distribution changing characteristics,which put forward higher requirements at mining technology. Utility pattern mining is a datamining under the premise of utility conditions. Therefore, how to mine the utility pattern is avery practical significant subject in the data streams.This paper continues the research about the frequent itemsets mining,the data streammining and the utility pattern mining.The MMIDS method solves the issue that mining themaximal high utility under the premise of utility pattern in the data streams. A compressionstructure called the maximal utility itemsets tree is designed in this paper, while implement itin landmark window model and the sliding window model. And then, in order to improve theefficiency of tracing the maximal utility itemsets tree, we propose a bottom-up trace strategyto reduce the number of nodes. Finally, as the real-world resources limited, MMIDS methoduse a pruning strategy for the maximal utility itemsets tree in ord to adapt to the data streammining conditions.The end of the paper is the experimental section. For evaluating theperformance of the MMIDS method, this papaer implements the comparative experiments atthe aspects of the running time, the space overhead, the scalability and the effectiveness of thebottom-up trace strategy, and then comparing with the the traditional methods MHUI-TIDmethods and THUI-Mine method. The experimental results reveal that there are significantimprovements in terms of the running time, the space overhead and the scalability.Furthermore, the MMIDS method use an effective pruning strategy, so that it can meet theneeds of the real world data stream mining under the condition of limited resources,.