Parallel Computation of a MMDBM Algorithm on GPU Mining with Big Data

Abstract

Big data is the collection of data sets which are large and complex in nature. It contains structured and unstructured types of data. For example, Financial Services, Retail, Manufacturing, Healthcare, Social network (Twitter, Fackbook, Linkedin and Google), Digital pictures and Videos. To extract useful data from big data, several classifiers like SLIQ, SPRINT, MMDBM are used. Among this one of the fast classifier is the Mixed Mode data Based Miner (MMDBM) using Graphical Processor Unit (GPU) mining. This classifier describes the outline of parallel computing with high performance, using radix algorithm for multicore GPUs, by taking a program presented by Compute Unified Device Architecture (CUDA). The classifier can deal with both categorical and numerical attributes in a simple manner. The classification method handles big data with huge number of attributes by taking it from the medical data base. This can be parallelized on GPU to get high-speed and better performance than CPU-Radix sort algorithm. We proposed the parallelized Radix sort algorithm on GPU computing using CUDA platform developed by NVIDIA Corporation. In this chapter, we discuss the performance of fast classifier method and radix algorithm to relate the processing time of MMDBM, SLIQ CPU with GPU computing and computed acceleration ratio (Speed-up) time. Also, The classifiers [SLIQ, SPRINT, MMDBM] are evaluated and compared with CPU and GPU. GPU provides quick and accurate results with least processing time and supports real time applications.