Parallel Ensemble Techniques for Data Mining Application

Abstract

Data mining is a powerful technique to extract hidden predictive information from large datasets. Classification is a very popular application of data mining. The goal of classification is to build a classifier with high accuracy and low cost. Since most data mining datasets are very large, using small subsets of the dataset can speedup data mining tasks. It is possible to generate a family of predictors using different subsets of training dataset and combine those predictors to achieve higher accuracy with low cost. This is the basic idea of ensemble techniques. Variants of ensemble techniques include bagging, AdaBoost, arcing etc. In this research work, parallel bagging and parallel AdaBoost with support vector machine as a base classifier are implemented with the support of the parallel hardware. The NSL-KDD datasets are used to examine the essential parameters for parallel ensemble techniques such as sample size, number of iterations, number of processors, and threshold. Experiments are conducted to demonstrate that ensemble techniques can be effectively parallelized. The parallel ensemble techniques will provide more accurate results.