Hellinger distance-based stable sparse feature selection for high-dimensional class-imbalanced data

摘要

Feature selection has recently gained considerable attention in class-imbalance learning due to the high-dimensionality of class-imbalanced data across many scientific disciplines [ – ]. To date, a variety of feature selection methods have been proposed to address high-dimensional data. However, only a small number of them are technically designed to handle the problem of class distribution under a class-imbalance setting [ – ]. Thus, performing feature selection from class-imbalanced data remains a challenging task due to the inherent complex characteristics of such data, and a new understanding or principle is required to efficiently transform vast amounts of raw data into information and knowledge representation [ ].