Certain investigations on big data classification algorithms utilizing extreme learning machine and online feature selection techniques

Abstract

With the arrival of big data learning has become unavoidable in many fields such as image processing information retrieval machine learning and pattern recognition The main difficulties in learning from big data consist of three aspects First it is hard to finish computation on a single computer within a tolerable time Second the high dimensional features degrade the performance of the learning algorithm Finally the transformation of learning concepts is hard to realize due to dynamic increase of data volume In order to overcome these difficulties parallelization of classification algorithms is needed One classification algorithm where parallelization is applied is known as Extreme Learning Machine Tree ELM Tree model The ELMs in the ELM Tree have the advantage of fast training speed and great potential to learn from big data Consequently to have a parallel ELM Tree model computation of the learning components needs to be carried out in a parallel manner In this thesis various techniques are proposed to improve the performance measures of this parallelized classification algorithm Firstly optimizing and parallelizing of ELM Tree model for big data classification known as OPELM Tree model is studied For parallelization purpose MapReduce paradigm is utilized Swarm intelligence newlinebased algorithms such as Genetic Algorithm GA Particle Swarm Optimization PSO algorithm and Firefly Algorithm FA is employed as the optimization techniques newline newline