ELM-Based Imbalanced Data Classification-A Review

Abstract

Imbalance issues occur in Machine Learning (ML) when there is high distortion in the class distributions. A great challenging task in ML is the imbalance of data classification. It is because most classification methodologies tend to bias toward the majority class even though high importance is given to the minority class. To enable its stable operation, many techniques are utilized recently that are still in use for classifying imbalanced datasets efficiently. Owing to the assumption with balanced class distribution or equal misclassification, the prevailing learning algorithms are prone to favor the majority class when handling complicated classification issues with skewed class distribution. The most prominently adopted technique to deal with data having imbalance class distribution is Extreme Learning Machine (ELM). Unwanted class boundaries as of data with unbalanced classes may be learned by ELM similar to various other classification algorithms. Grounded on the kernel utilized, elevated weighted ELM, active learning-centered techniques, etc, an augmentation in the ELM framework is done for efficient imbalanced classification. Regarding ELM approaches, the latest studies for imbalance classification are studied here. Finally, regarding G-Mean, Accuracy, and Imbalance Ratio (IR), the research studies’ performance was analogized.