Empirical Study of Enhanced Sampling Schemes with Ensembles to Alleviate the Class Imbalance Problem (original) (raw)

2018

Abstract

Classification of an imbalanced dataset is sub-optimal as traditional classifiers are biased towards the majority class completely ignoring the minority class. However, this minority class is a class of interest and should not be ignored as its misclassification cost is higher. This research is aimed at identifying and treating imbalance dataset. This study proposes five SMOTE (Synthetic Minority Oversampling Technique)-based enhanced data sampling schemes with both homogeneous and heterogeneous ensembles to alleviate the class imbalance problem. Waikato Environment Knowledge Analysis (WEKA) filter library was extended with these enhanced sampling schemes for pre-processing of the datasets before their classification. Real life datasets collected from different domains in Nigeria were used for its implementation and Receivers’ Operators Characteristics Area Under Curve (ROC_AUC) and Performance Loss/Gain metrics were used as evaluation metric for these schemes. SMOTE300ENN, one of t...

Sakinat Tijani -Folorunso hasn't uploaded this paper.

Let Sakinat know you want this paper to be uploaded.

Ask for this paper to be uploaded.