Research On Algorithms For Mining Frequent Patterns In Data Streams

Recently, data streams have been widely used in network monitoring, sensor network data analysis, Web-click data stream analysis, financial trade and business transactions.Data streams are continuous, fast, unlimited, unknown, so traditional technology of data mining is not suitable to data stream mining. Analysis and mining data stream has been a popular research.Frequent pattern mining is one basic research of data stream mining. This dissertation first introduces the definitions and background of frequent pattern mining in data stream. Then it introduces and analyses the states of domestic and foreign researches.At last, it gives a detailed introduction of the basic knowledge and technical.Due to existing algorithms of frequent pattern need to deal with large search space and save intermediate results, this dissertation proposes algorithms of mining maximal frequent patterns in data stream with sliding window. First, Bitmap is used in this algorithm to deal with the streaming data. Second, when mining, we adopt depth first to find maximal frequent itemsets.Besides typical pruning strategies, we develop a new pruning strategy corresponding to the parent equivalency pruning to prune. Third, index structure is used to store the maximal frequent itemsets, which can speed up the speed of superset test. The results of the experiment prove that the algorithm is effective in time and memory.When determining the minimal support, it requires prior experience. This dissertation proposes a Top-k frequent itemsets algorithm without minimal support in sliding window, called MTKFI. The algorithm is based on Apriori ideas, and use bitmap information to get the frequent itemsets repeatedly. At the same time, it stores the frequent itemsets in a two indexes link. Finally, it outputs the most frequent k-frequent itemsets.