Mining Method And Application Of Strong Jumping Emerging Patterns Based On NSJEP-list

The Strong Jumping Emerging Pattern,which is proposed in recent years,is a kind of contrast patterns with the good distinguishing ability.The classifier based on SJEP can achieve a higher accuracy,so SJEP becomes an important research direction in classification research gradually.Because most of the current mining algorithms of SJEP are based on the data structure,tree,the problem of large consumption in time and space is frequently caused.Meanwhile,SJEP cannot show a good ability in classification in the imbalanced datasets,especially for the positive class.In this paper,after studying the mining and classification algorithms based on SJEP,the main work is as follows:The problem of large consumption in time and space is caused by the tree-based SJEP mining algorithm.In order to solve this problem,a new data structure named NSJEP-list is proposed.Based on this data structure,a new SJEP mining algorithm called PPSJEP is proposed.The algorithm constructs a PPSJEP-tree according to the training dataset firstly and then gets the individual items' NSJEP-lists to replace the PPSJEP-tree.Finally,the algorithm uses the intersection of NSJEP-lists to get the longer itemsets,NSJEP-lists which can be used to mining strong jumping emerging patterns.Experiment results show that the mining algorithm based on NSJEP-list can effectively filter out the redundant JEPs,and the mining efficiency is higher than the traditional SJEP mining algorithms which are based on tree.In the imbalanced datasets,since the number and the support of the contrast patterns mined from the positive class are less than the negative class,the classification accuracy of the positive class may be lower than normal.In order to solve this problem,an extended SJEP is proposed,which can mine some additional contrast patterns with good distinguishing abilities from the positive class.Based on the extended SJEP,to solve the problem of the traditional SJEP-based classification algorithms being conductive to the negative class,an improved classification algorithm which introduces a weight to the final score in classification is proposed.Experiment results show that the extended SJEP and the improved classification algorithm can achieve a higher classification accuracy especially in the positive class in the imbalanced datasets.