Research On Mining Sequential Rules Based On Negative Sequential Patterns

Negative sequential rules(NSRs)take non-occurring events into consideration on decision-making to analyze the potential meaning of data from a new perspective,which has great research significance and make up for the deficiency of positive sequential rules(PSRs)on decision-making.However,the existing NSRs mining algorithms are very limited,and there are many problems under solved.The problems are as follows:(1)due to negative sequential patterns(NSPs)do not satisfy the downward closure property,the sub-sequences of a frequent NAP may not be frequent and it would be unable to calculate the confidence of the corresponding NSR.At the same time,the confidence of some NSRs may be greater than 1,which makes it difficult for users to set the confidence threshold;(2)owing to it is the first time to mine NSRs from NSPs,we don't know which form of NSRs are reasonable.(3)mining NSRs from NSPs will generate contradictory rules.How to prune these contradictory rules to ensure the rules can be used for decision-making? This paper researches on the above problems,and propose a algorithm nsp Rule to discover NSRs from NSPs and a contradictory rule pruning algorithm ASR which based on contribution and correlation coefficient.The main contributions of this paper are summarized as follows:In order to address the first two problems,we propose an effective algorithm to discover NSRs from NSPs--nsp Rule.The algorithm first determines whether the antecedent or the consequent of the rule is a frequent pattern,and then solves the problem that the confidence of the rule cannot be calculated by deleting the infrequent pattern.For the problem that the confidence of some rules are greater than 1,nsp Rule algorithm normalizes the confidence of rules to ensure the confidence range of rules is within [0,1].In addition,in order to ensure the form of generated rules is reasonable,nsp Rule algorithm takes the correlation of rules into consideration to effectively avoid the generation of rules with unreasonable form.To the best of our knowledge,nsp Rule is the first algorithm to discover NSRs from NSPs.Experiments on real and synthetic datasets show that nsp Rule algorithm can effectively mine NSRs from NSPs.In order to solve the third problem,we propose an algorithm based on contribution and correlation coefficien--ASR.Firstly,this algorithm take the concept of contribution into account to consider the internal relationship between the antecedent and the consequent of the rule,so as to prune the contradictory rules whose internal relationshipdoes not meet the conditions.Based on that,the ASR algorithm also takes the correlation coefficient into account to prune the contradictory rules between the antecedent and the consequent of the rule.At the same time,ASR algorithm also combines contribution,correlation coefficient and Wu's pruning framework,which enables ASR algorithm to prune contradictory rules w.r.t.more comprehensively aspect including rule correlation,rule internal relationship,rule support and confidence.So as to ensure that the rules can be used for decision-making.The experimental results prove that the ASR algorithm is very effective.In addition,this paper also applies nsp Rule algorithm to the consumption data of university students',and obtains some meaningful rules.The results show that students' academic performance is significantly related to whether they have breakfast on time.