The Research And Application Of Mining Frequent Patterns Over E-Bank Data Straems

With the development of information technology, achieving information and knowledge from the explosive, highly dynamic data has become the core capability of an enterprise even a country to taking the leadership in an industy. Data mining is just the artificial intelligence technology to solve this kind of problem. But nowadays some new application like fraud monitoring in securities trading and risk control of financial transfers, sensor detection and precautionary of credit card fraud require more time efficiency even most time need real-time reaction. For this reason, "data stream" has been proposed in the last century. Different from traditional database technology, the data stream is continuous, orderly, infinite, and queries on data streams require highly real-time feedback. It's usually allowing only one or very few number of scans over data stream. To obtain useful information from vast amounts of data stream with limited system resources has brought new opportunities and challenges for data mining.In this paper, we researched the technology of mining frequent patterns over e-bank data stream and other related technologies. We focuses on the following issues:existing data stream data structure model and algorithms for mining frequent pattern over data stream; structural characteristics of financial online bank data stream and its data model; improved Lossy Counting to achieve more dynamicity; developed a new algorithm named MFS-HT to mining frequent items over e-bank data stream based on sampling and hash-table; the data structure of e-bank data stream, design and develop e-bank data stream mining system for frequent pattern. The research concept and innovative work of this paper including:First of all, we had researched the theory of data stream mining and its data structural model. And then we summarized the current outcomes of the latest researching in this domain. With the studying of e-bank data stream we summed up the concept and characteristics of e-bank data streams, conclude that e-bank data stream has more features like sequential, multi-semantic, conflicted, and magnific than general data stream.Then a new data stream model of e-bank data stream-time-series turnstile model was proposed. Then, we have researched and analysised of existing algorithms for mining frequent patterns over data stream. Based on time-series turnstile model, TTLC (Time-series Turnstile Lossy Counting) algorithm was proposed. It improved the dynamicity of Lossy Counting.We also designs and implements a hash table based algorithm for mining frequent patterns named MFS-HT, and proved by experiments that the time and space efficiency of MFS-HT is superior to existing algorithms.Finally, the above algorithms had been implemented. And a frequent patterns mining system over e-bank data stream had been designed and implemented. With the experiment on this system and analyzing on the experimental results, we concluded that the system has high accuracy and time efficiency.