The Algorithm Research Of Association Rules Mining

Comparing with the traditional statistic and query methods, Data Mining concerns multiple subjects, congregating some research results of artificial intelligence, pattern recognition, database, machine learning and management information system, etc. Data mining is a newly-established frontier subject. It is being used extensively and its application future is bright.This thesis studies and analyses the data mining technique systematically and deeply, especially for association rules. The main contents are listed as follows:Research and analysis of association rules. Based on the research of the existed document of association rules, in this dissertation, the basic concepts and property are introduced roundly and its typical mining algorithms and these algorithms' basic ideas are summarized,analysed and studied.For the question solving in the dissertation, the classical algorithms-the Apriori algorithm and FP-growth algorithm are analysed and studied. All kinds of optimized techniques which are designed to promote the Apriori algorithm's efficiency are also studied here . All of the above rationally establish the necessary premise for the improved algorithm's proposition and construction.Design, analysis and research of the improved algorithm for Apriori. Based on the work above, in the dissertion, the improved algorithm for Apriori is put forward. This algorithm mainly takes into consideration the bottleneck problem of frequent itemsets generation, it improves the Apriori algorithm by reduce of scan times of transaction DB etc. Successively the process of mining association rules using the improved algorithm is illustrated through an instance.Experimental results of the improved algorithm. Based on the synthetic data using Possion distribution funtion and exponential distribution function, the performance of the improved algorithm and the comparison among the improved algorithm, Apriori algorithm and FP-growth algorithm is experimented. Finally the results of the experiments are analysed. All of the experiments reveal good performance of the improved algorithm.