Arm++: A Hybrid Association Rule Mining Algorithm

Most of the approaches for association rule mining focus on the performance of the discovery of the frequent itemsets. They are based on the algorithms that require to transform data from one representation to another, and therefore excessively use resource and incur heavy CPU overhead. This paper proposes a hybrid algorithm that is a resource-efficient and provides better performance. It characterises the trade-offs among data representation, computation, I/O, and heuristics. The proposed algorithm uses an array-based item storage for the candidate and frequent itemsets. In addition, we propose a comparison algorithm (CmpApr) which compares candidate itemsets with a transaction, a filtering algorithm (FilterApr) which reduces the number of comparison operations required to find frequent itemsets. The hybrid algorithm (ARM++) integrates filtering method within Partition algorithm [7]. Performance analyses from our implementation indicates that ARM++ has better performance and scales linearly.