Imbalanced data classification using random subspace method and SMOTE

Abstract

Class imbalance problem has attracted many attentions in recent years. When the available training sample size of each class is imbalanced, the directly established classification model will tend to allocate the testing sample into the majority class. A proper resampling method together with a power classifier is generally employed for dealing with this problem. Many multi-classifier ensembles have been shown to outperform single classifier in many experiments. Bagging and boosting are two most popular multi-classifier frameworks and have been applied to deal with the class imbalance problem. By observing that the sample information of the minority class is very limited and the small sample size (SSS) problem might decrease the performance of the classifiers, another powerful multi-classifier method called random subspace method (RSM) is introduced to deal with the class imbalance problem in this study. To evaluate the performance of different classifiers, a well-known resampling method called SMOTE is employed. The experiment results showed RSM has the best performance in most of the considered situations.

A Wide Scale Classification of Class Imbalance Problem and its Solutions: A Systematic Literature Review

“…Based on literature survey, the ensemble methods (Breiman, 1996;Joshi et al, 2001;Sun et al, 2012) like bagging and boosting are the two most popular multiclassifier frameworks and have been applied to deal with the class imbalance problem. Huang et al (2012) introduced a multi-classifier method called Random Subspace Method (RSM) to deal with class imbalance problem. Abolkarlou et al (2014) proposed a novel technique of the ensemble classification for handling the imbalance data using uses hierarchical clustering algorithm for determining the optimal layers.…”