Design And Study Of The Data Mining System Based On Sequential Database

Sequential pattern mining, which discovers frequent subsequences as interesting patterns in a sequence database, is an important and active research problem with broad application at present, including the analysis of customer purchase patterns and Web access patterns, the analysis of sequencing or time-related processes such as scientific experiments, natural disasters, and disease treatments, the analysis of DNA sequences, and so on- So, the research of the sequential pattern mining will strongly promote the related area researches.Users may repeatedly put forward many different support thresholds at random, if the running speed of the data mining system is very slow, users may not tolerate so long period. For improving user' s satisfaction in the system functioning, the system can save the sequential patterns into the knowledge base that gained in the system running, and speed up the system running. This dissertation put forward the structure of the sequential pattern mining system and the knowledge saved in the KB. The knowledge structure stored in KB including the minimum support threshold, sequential patterns and their support, these sequential patterns stored in KB in block based on their support. The dissertation proposed the method of creating KB, and when and how to update the KB.As to sequential pattern mining, mining algorithm is very important. Researchers are studying hard to develop efficient data mining algorithms at all times. Based on some sequential pattern mining algorithm, the dissertation analyzed the shortage of the MEMISP algorithm, and proposed an improved MEMISP algorithm. The improved MEMISP algorithm projects original database into the memory with frequent items, and removes those infrequent items, the algorithm avoids useless reading when the system is running. The paper proposed new data structure and processing method of the improved algorithm, this makes the new algorithm more efficient and easy to run.