文献类型：期刊文献

英文题名：Graph discriminative dynamic sampling AdaBoost for imbalanced cardiovascular disease diagnosis

作者：Li, Chuantao[1];Chen, Baoqin[1];Li, Sheng[1]

机构：[1]Guangdong Ocean Univ, Sch Math & Comp, Zhanjiang 524088, Peoples R China

年份：2026

卷号：297

外文期刊名：EXPERT SYSTEMS WITH APPLICATIONS

收录：SCI-EXPANDED(收录号：WOS:001565934500001)、、WOS

基金：We sincerely thank the anonymous reviewers for their time and great efforts in reviewing our paper, their valuable comments, suggestions, and advice helped us significantly improve the quality of our manuscript. This work is supported by the National Natural Science Fund of China (No. 12101138) , the National College Students Innovation and Entrepreneurship Training Program (No. 202510566026) , Students Innovation and Entrepreneurship Training Program of Guangdong Ocean University (No. CXXL2024148) , and Guangdong Ocean University Undergraduate Innovation Team Project (No. CXTD2023014) .

语种：英文

外文关键词：Cardiovascular disease; Class imbalance; Dynamic sampling; Discriminability LPP; AdaBoost variant

外文摘要：To tackle the challenges arising from class imbalance within cardiovascular disease (CVD) datasets, this study proposes a novel diagnostic model: Graph Discriminative Dynamic Sampling AdaBoost (GDDSAD). This model jointly optimizes class balancing, feature extraction, and the classification process within a unified AdaBoost architecture. Initially, the weight updating mechanism leverages instance confidence to prioritize boundary-proximal instances. Subsequently, dynamic undersampling and dynamic oversampling modules are integrated during the AdaBoost iterations in order to construct a more distinct decision boundary. Ultimately, the dynamically sampled dataset is dimensionality reduced using discriminability improving Locality Preserving Projection (LPP) to provide a more discriminative low dimensional manifold input to the base learner. Experimental results demonstrate that GDDSAD achieves an average increase of 0.208 in the G-Mean and 0.111 in the AUC on 4 CVD datasets compared to AdaBoost, and outperforms mainstream Boosting variants on 6 KEEL datasets, demonstrating its ability to accurately identify diseased instances in imbalanced CVD datasets. The code related to this study is available in the GitHub repository: https://github.com/ChuantaoLi/GDDSAD.