Research On Association Rule In Data Mining Based On The Interestingness

Data mining as a branch of artificial intelligence,gradually being applied to all aspects, from the primary to the current mass and extensive application, people spend a lot of effort on putting forward a lot of research methods, and develop the discipline.In the process of development, according to different data mining application, Data mining has respectively developed mining methods of cluster, classification, association rules.In the early,mostly research is concentrated in support- confidence threshold, but in the process of practical application,may produce many false misleading rules.Therefore,generation of association rule considering the degree of support and confidence threshold,besides, still need to introduce interestingness, to check results whether have value or not.At present, people have proposed different interestingness measures, but there is not widely accepted interestingness measure which still need to be improved.Therefore, to cover the shortages of the traditional support-confidence,combined with analysis of some existing interestingness measures, puts forward a new interestingness measure, with examples proving the effectiveness of the proposed measures.At the same time,using interestingness measure to mine valuable positive and negative association rules, is faced with explosion problems of the negative association rules. Although many people have proposed algorithms on mining association rules in positive and negative correlation from different aspects, there are still many problems to resolve in reducing explosivity of the negative association rules.This paper from the perspective of reducing the infrequent items, combined with new interestingness measure, introduced the maxsup,proposes an algorithm for mining positive and negative association rules,and proved the algorithm could reduce the negative association rules or mislead the rules through the experiments on Mushroom data. Due to the negative association rules require not only meaningful but also should be readable, so this paper proposes a new negative association mining algorithm,which not only can greatly reduce the irrelevant and misleading rules, but also has good readability through the experiment compared with the existing algorithm.In the end, this paper applys the improved algorithm to research the stock price linkage, studying the stocks positive and negative correlation with each others, on the one hand, the improved algorithm is verified practically through the example;on the other hand, in the study also gains some valuable association rules.