Parallel FIM Approach on GPU using OpenCL

Abstract

In this paper, we describe GPU-Eclat algorithm, a GPU (General Purpose Graphics Processing Unit) enhanced implementation of Frequent Item set Mining (FIM). The frequent itemsets are extracted from a transactional database as it is a essential assignment in data mining field because of its broad applications in mining association rules, time series, correlations etc. The Eclat approach is the typically generate-and-check approach to obtain frequent itemsets from a database with a given minimum support threshold value. OpenCL is a platform independent Open Computing Language for GPU computation. We tested our implementation with an Radeon Dual graphic processor and determine up to 68X speedup as compared with sequential Eclat algorithm on a CPU. In order to map the Eclat algorithm onto the SIMD (Single Instruction Multiple Data) execution model, an array data structure is used to represent the input database and standard dataset is converted to the vertical data layout. In our implementation, we perform a parallelized version of the candidate generation and support counting phases on the GPU. Experimental results show that GPU-Eclat consistently outperforms CPU-based Eclat implementations. Our results reveal the potential for GPGPUs in speeding up data mining algorithms.