Searching the certainties from the uncertainty: A knowledge enhancement model for imbalanced medical data
Jie Ma, Wenjing Sun, Zhiyuan Hao · Information Processing & Management · 2024
Medical data typically encompass a multitude of features, which contain vast hidden knowledge and also exhibit deep uncertainties. How to search the valuable features is a significant consideration for discovering high-quality medical knowledge. To this end, we introduce the optimization method to conduct feature selection and propose a knowledge enhancement model for imbalanced medical data (IMD-KEM). Specifically, IMD-KEM is featured with two parts: (1) searching for potential target features; (2) knowledge discovery based on feature subset. In part (1), we introduce balancing mechanism (BM) and pole migration mechanism (PMM) into sparrow search algorithm (SSA) thereby propose an enhanced sparrow search algorithm (ESSA), which possesses both better exploration and exploitation capabilities to identify target features. In part (2), we initially enable ESSA to conduct binary transformation by designing a binary transformation mechanism (BTM), leading to the proposal of the binary enhanced sparrow search algorithm (BESSA) for obtaining a suitable feature subset. Subsequently, we integrate SVM with BESSA and SMOTE to discover knowledge from re-balanced feature subset, where SMOTE is utilized to re-balance the distribution of feature subset. Moreover, this paper conducts two sub-experiments to test the validity of IMD-KEM: one focusing on optimization algorithms and the other on feature selection, where various baseline swarm intelligence optimization algorithms and baseline feature selection methods composed of these algorithms are compared. This paper employs CEC 2014, CEC 2017, and CEC 2019 to verify the superiority of ESSA in experiment I, while adopting 6 real imbalanced medical datasets and 6 evaluation metrics to test the validity of IMD-KEM in experiment II. The experimental results demonstrate that IMD-KEM performs significantly in discovering hidden high-quality medical knowledge.