Research On Positive And Negative Association Rules Mining Based On Multi-confidence Threshold

Association rules mining which reveals the interesting correlations among itemsets of datasets is an important research direction in the field of data mining,and has significant theoretical value,together with broad application prospects.However,most of association rules mining research works only focus on the positive correlations among itemsets,and the negative correlations among itemsets hidden in the datasets did not attract much attention.In practical applications,the negative correlations among itemsets can provide more valuable decision information for decision-makers.Therefore,the study of positive and negative association rules mining has great practical significance.In this thesis,the basic theory of positive and negative association rules mining is discussed,the key elements of designing the effective positive and negative association rules mining algorithm are analyzed deeply,and the shortcomings of the existing mining algorithms are summarized.The existing algorithms for mining positive and negative association rules based on multi-confidence threshold are difficult to effectively set up multiple confidence thresholds to control the number of boring rules and extract high reliability rules.And they also easily omit some interesting association rules during the mining process.In this regard,the algorithm for mining positive and negative association rules based on multi-confidence threshold is analyzed and researched deeply in this thesis,and some research results are shown as follows:Firstly,the characteristics of the confidence of rule changing with the support of itemsets of rule are analyzed systematically with itemset correlation.According to the change characteristics,a new two-stage confidence threshold setting method of positive and negative association rules(called TCTPN)is proposed.The results of theoretical analysis and experiment show that the new method can not only ensure that the extracted association rules are effective and interesting better,but also reduce the number of association rules with low credibility significantly.Secondly,a new algorithm for mining positive and negative association rules based on correlation measure Kulc and TCTPN(called PNARKT)is proposed.Based on the theory that the positive itemsets of both the antecedent and descendant of interesting association rules are frequent itemsets,the proposed algorithm generates strong positive and nagetive association rules by analyzing the correlation between two disjoint frequent itemsets in the transaction database,reducing the omission of interesting rules.Meanwhile,Kulc and TCTPN are introduced into the new algorithm to ensure that the rules extracted by PNARKT algorithm are interesting and credible.Theoretical proof and experimental comparison results show that PNARKT algorithm can not only avoid the omission of interesting positive and negative association rules better,but also extract interesting positive and negative association rules in transaction database with a large number of null transactions and some itemsets which have unbalanced implication relation effectively.