Ensemble Methods for Improving Classification Accuracy On Heterogeneous Datasets

Abstract

Classification algorithms are the algorithms which mine huge events, patterns newlineor entities to classify multiple attributes or patterns based on their similarity or newlinerelationship. Classification systems have been playing decisive role in different newlinesectors such as finance, business, defence systems, scientific research, healthcare, newlineindustrial activities monitoring and control etc. In fact, classification systems have newlineemerged as an inevitable component for the contemporary decision-making systems. newlineAs stated, classification systems have irreplaceable significance towards newlinedecision making purposes, different approaches such as regression techniques, pattern newlineclassifiers like support vector machine, neural networks etc have been developed. newlineHowever, majority of such classification methods have their respective limitations newlinelike convergence, local minima issues etc. In addition, different classifiers have the newlinedifferent efficiency even for the same data. On the other hand, expecting same newlineefficiency over different and more complex data structure can t be fulfilled optimally newlinein all cases. It signifies the need of a robust and efficient classification system which newlinecould accomplish a generalizable and convincing solution for different classification newlineproblems. As an alternative, in the last few years, ensemble learning methods have newlineemerged as a potential solution. Ensemble learning is a state-of-art classification newlinesystem which embodies multiple (independent) classifiers (also called as base newlineclassifiers) in such strategic manner that it yields optimal performance or accuracy newlineeven over complex and large-scale data. In other words, ensemble learning is one of newlinethe most efficient classification systems, which exploits efficacy of the different base newlineclassifiers to solve certain classification problem. Ensemble learning approaches have newlineproved their efficacy over conventional classification systems; however upsurging newlinedata complexity, heterogeneity and the selection of the best ensemble classifiers have newlinealways been the challenge for industr