An adaptive algorithm for frequent pattern mining over data streams using diffset strategy

Frequent pattern mining using sliding window over data streams is commonly used due to its wide applicability. Determining suitable window size and detection of concept change are the major issues and can be addressed by having flexible window based on amount of changes in frequent patterns. For mining frequent patterns over data streams, vertical mining algorithms can be used. However, in these algorithms, size of transaction identifiers (tidsets) and the time for computation of intersection between tidsets is large. Moreover, presence of null transactions does not contribute any useful frequent patterns. A new algorithm called recent frequent pattern mining based on diffset with elimination of null transactions (RFP-DIFF-ENT) over data streams using variable size window is proposed. It stores difference of tidsets and eliminates null transactions which minimise memory and mining time. Experimental results show that proposed algorithm saves computation time, memory usage and minimises the number of frequent patterns.