Publication: Minimizing Classification Errors in Imbalanced Dataset Using Means of Sampling
Date
2021
Authors
Khan I.
Ahmad A.R.
Jabeur N.
Mahdi M.N.
Journal Title
Journal ISSN
Volume Title
Publisher
Springer Science and Business Media Deutschland GmbH
Abstract
Classification, a significant application of machine learning, labels each instance of the dataset into one of the predefined classes. Problems occur when the number of instances in the classes is not uniform. The exceptional lyuneven class distribution gives rise to class imbalancing issues which tend to demote the overall performance of the classifier. A set of data-level algorithms are available which are applied to adjust the class distribution. The class imbalancing emerges frequently in datasets from educational domains where the number of students with unsatisfactory performance general appears in low number comparing to the students with satisfactory outcomes. This paper applies a set of data-level sampling algorithms over a dataset taken from an educational domain. It underlines the consequences rising from classification with imbalanced dataset. This research confirms that a classification model achieving higher accuracy may not appear effective in correct identification of instances in minority class. Classification with an imbalance dataset may produce low recall, precision and F-Measure for classes with lower number of instances. The performance of classification model improves with application of data level algorithm. However, it highlights the supremacy of oversampling algorithm over undersampling algorithms. � 2021, Springer Nature Switzerland AG.
Description
Classification (of information); Learning algorithms; Students; Class imbalance; Data level; Over sampling; Performance prediction; SMOTE; Spread subsampling; Student performance; Student performance prediction; Under-sampling; Machine learning