Abstract:
Nowadays, consumers behavior analysis is a crucial issue in competitive
business. There is a need to know consumers behavior including changes of
consumers behavior. This leads to an emergence of mining interesting itemset. Since
the first proposed to discover emerging patterns (EPs) which can help to know trends
and differences on occurrences of itemsets in the term of frequency. However, mining
EPs only considers changing on frequency of occurrence of itemsets which may not
sufficient to express change on regularity or irregularity of itemsets in several real-life
applications such as tracking changes of buying behavior, monitoring changes of effects
on patients after using medicines, observe change in travelers preferences of hotel
business and so on. To solve the above limitation, we propose to (i) Discovering
interesting itemsets based on change in regularity of occurrence. An efficient singlepass algorithm based on pattern-growth concept named MICRO. A tree-based structure
called ICRO-tree is also designed to efficiently maintain candidate itemsets with their
essential information. A property used for pruning search space is also introduced in
order to reduce resource usage during mining process. However, this approach
overwhelming of generated results and difficulties to the users. Hence, it is helpful to
avoid this which can help users to be more efficient to look for interesting information
and/or knowledge from these itemsets. Therefore, to address this issue, we propose
to (ii) Mining regular itemsets with interesting changes in regularity of occurrence in
order to generate a compact set of results based on the user-given regularity and
change thresholds. An efficient single-pass algorithm named RICROM and a new
interval word segment structure called NIWS are designed to efficiently mine such
itemsets and maintain occurrence information of each itemset. Experiments were done
in real and synthetic datasets. The results illustrate the efficiency of runtime, memory
usage and the number results of discovered.