The Research On Application Of Multiobjective Association Rules Mining Algorithm Using Swarm Intelligence

Association rules mine is a very important research topic in the data mining which is widely used in various fields. It can not only test the long formatted knowledge model,but also can find the hidden new laws. Effectively discovering, understanding and using association rules is an important means to finish the data mining. The research of association rules mining in large data environment has important theory value and practical significance.Apriori algorithm is the most influential mining frequent items algorithm. FP-growth algorithm is efficient classical algorithms for association rules mining. The two traditional algorithms have three weakness:(1) the candidate set is too much, algorithm must spend a lot of time to deal with the candidate itemsets;(2) May repeatedly scans the database;(3) algorithms only conside the fixed minimum support and minimum confidence. With the various evaluation measures introduced, the association rules mining is not a single objective problem but a multi-objective optimization problem. The Assocation rules mining algorithm based on genetic algorithm is a classical multi-objective algorithm, but there are two problems.(1) low efficiency;(2) the association rules of mining is a set which contains a large number of association rules, unable to further process, and users often don’t know the which of the rules is most in need. The research work in this paper focuses on solving these 3 problems with different methods.At first, we propose a new binary Bat search algorithm(BBA) and adopted a new binary mathematical model, which not only improve the convergence but also ehance the diversity of results. Then the new Association rule mining algorithm based on BBA is proposed, experimental results show that the global search ability and convergence of BBA is higher than the association rule mining algorithm based on BPSO. At the same time it can dig the rare mode association rules which the Apriori algorithm can’t, so it makes up for the weakness of the traditional algorithm.Secondly, because the frame of confidence and support can not be good to evaluate the association rule, the researchers introduce the multiple measures to evaluate the association rules. So the association rules mining algorithm is a multi-objective optimization problem, we can take confidence, support, lift, interestness as the target algorithm to optmize. In order to solve the multi-objective problem, we put forward a new multi-objective binary bat algorithm(MBBA). Because the multi-objective algorithm produces the Pareto solution set which contains a number of rules, how to judge the rule which is more suited to the user? We put forward the Degree of Similarity(Deg Sim) method. This method is applicable to comprehensively evaluate association rules and helps the user to choose the best association rule in the Pareto solution set. The experimental results show that based on MBBA and Deg Sim mining association rules is superior to the single objective BPSO and BBA algorithm.Thirdly, time series, as an important data type, has its own Partieularity besides the universal Characteristics of data sets. If we can use the advanced data mining technology to analyze the useful patterns.in the data sets, this exploratory study clearly is of great importance both in theory and practice. Combining characteristics of time series association rule and the requirements of practice, on the basis of above reseach, we design a time sequence association rules mining algorithm model based on MBBA and use the new algorithm model to explore the application in the analysis of steel prices, by the empirical test to further vertify the availability, effectiveness, useful of the algorithm.The research work in this paper, the classical association rule mining algorithm is analyzed in different angles and using different solution to solve the weakness of it. Faced with the hot research currently, Time series association rules mining, we puts forward a new solution. Faced with the hot research currently, the research content of this article has certain theoretical significance, especially the part time series association rules mining applied to steel prices data sets has good practical value.