Mining Association Rules Over A Stream Sliding Window

As a product of the information age,data stream is widely used in various fields ofsocial life.Data stream contains a wealth of knowledge.Particularly,the associationrules between mass of data is an important basis for decision making in prediction andonline analysis systems.Most existing researches are focusing on transaction datamodel;few are mining the association rules between elements occurring in data stream.In special environments,there are always dependencies between the data.Due to thecharacteristics of real-time data,people are more interested in the information of recentdata than that of the old.In order to mining association between the most recent data elements,we proposean MARSW algorithm to report the association rules under the sliding window model.The algorithm splits the sliding window into a series of sub-windows,maintains thesynopsis data structure by the operations of sub-windows.The experimental resultsshow that the proposed method can retrieve all the association rules between largeamounts of data under the sliding window at limited space cost within the given errorband,and have high efficiency and excellent scalability.Taking into account the data variability and flow of non-controllable,thealgorithm need data stream management system(DSMS)to pre-process the datastream.Since the DSMS has no widely accepted standard,we propose a Federatorstructure to support various DSMSs.By creating a built-in operator or taking over theinput and output,association rule mining algorithm can be quickly integrated intoDSMS under the unified interface model.