Mining association rules with new measure criteria
Yong Xu, Sen-Xin Zhou, Jin-Hua Gong · 2005
Nowadays, association rules mining from large databases is an active research field of data mining motivated by many application areas. However, there are some problems in the strong association rules mining based on support-confidence framework. Firstly, there are a great number of redundant association rules generated, then it is difficult for user to find the interesting ones in them; secondly, the correlation between attributes of given application areas is ignored. Therefore new measure criteria, which are Chi-Squared test and cover, should be introduced to association rules mining, and the more important aspect is the use of Chi-Squared test to reduce the amount of rules. In the paper, the Chi-Squared test and cover of measure criteria would be used to association rules mining for removing the itemsets, which are statistic independent, while frequent itemsets or rules are generated. Thus the number of patterns (including itemsets and rules) itemsets decreases, and it is easy for user to capture the more interesting association rules. The experimental results demonstrate that the Chi-Squared test is effective on reducing the amount of patterns via merging support and cover constrain. Pattern selection based on Chi-Squared test can eliminate some irrelevant attributes, and the efficiency and veracity of mining association rules are improved.