ScadiBino:...

Abstract

Current data mining algorithms are impractical for huge amounts
of data because they are time consuming and therefore
inefficient. Association rule mining is one of the most famous
data mining algorithms. Many parallel and distributed methods
have been proposed for association rule mining. However, these
methods are not suited to big data for a number of reasons, such
as improper data location, data skewness, lack of load balancing,
lack of support for generalized association rule mining, and lack
of an obvious method for rule extraction. The MapReduce-based
architecture is a parallel and distributable solution for association
rule mining. To improve the performance of MapReduce,
proposed methods for association rules need to be customized.
The performance of iterative algorithms in MapReduce
architectures may not be optimum. Two main issues affect the
performance of MapReduce architectures: data placement and
network traffic. In this paper, a scalable and distributable
binominal association rule mining method (ScaDiBino ARM) is
proposed. This method converts input data items to binominal
format to take advantage of scalable and distributable attributes
of MapReduce structures. The proposed method was evaluated
by applying it to real traffic data of a mobile operator to enable it
to recommend values added services (VAS) to its customers. The
results show that the rule extraction time improved significantly
after applying the proposed rule mining method.