Performance Evaluation of Binary and Multi-Class Dataset using Ensemble Classifiers

Abstract

Now a day’s digital data is dominating the entire globe. Every day it is increasing as it is generated from various domains such as social media, healthcare, education, banking etc. as well as through smart devices, IoT Devices etc., which we call these days Big Data. Due to the availability of Big Data, ensemble machine learning methods represent an attractive approach that can be used to deal with mining large and complex datasets. Ensemble models are now standard in Big Data Mining due to the fact that combining multiple classifiers together on a large dataset can often produce a much powerful classifier. The main principle behind this approach is that when weak classifiers are correctly combined, we can get better results. With this research paper, we have made a noble attempt to compare the performance of proposed methodology with existing research study using the various data mining classifiers. This study also compares the performance of basic data mining classifiers with the ensemble classifiers to solve the two major classes of classification problems: Binary-Class and Multi-Class in terms of accuracy. Ensemble approaches are implemented here to improve the performance of simple models and reduce overfitting of more complex models. The experimental results show that for the Multi-class classification task, Bagging performs well in comparison of Binary-class. But for Binary Class dataset, it is found that in most of the models, basic classifiers perform better than the ensemble classifiers. Moreover, it is observed that the Bagging performs well for all types of training-testing splits of the datasets.