We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
An optimized AdaBoost algorithm with atherosclerosis diagnostic applications: adaptive weight-adjustable boosting.
- Authors
Wang, Sensen; Liu, Wenjun; Yang, Shuaibin; Huang, Hui
- Abstract
In this study, a new boosting algorithm was proposed based on the traditional AdaBoost algorithm to better address classification problems. While preserving the core idea of AdaBoost, the proposed algorithm introduced a parameter-controlled adjustable function to further regulate the update strength of sample weights, enhancing its robustness and improving its ability to handle challenging classification tasks. The rationality and feasibility of the improved algorithm were theoretically proven, and the optimal formula for weak classifier weights was provided. Experimental results from numerical simulations demonstrated that the proposed method significantly improved performance compared to the traditional AdaBoost algorithm on datasets containing noise features and class imbalance, especially under sample insufficient conditions. Furthermore, when combined with the RFE-RF feature selection technique and applied to real atherosclerosis data, our algorithm achieved optimal performance while retaining six important features: total cholesterol (TG), low-density lipoprotein (LDL), fasting blood glucose(FBG), white blood cell count (WBC), uric acid (UA), and left carotid artery early systolic pulse wave velocity (LCCAES). Notably, our algorithm outperformed the AdaBoost algorithm across all evaluation metrics, including accuracy, precision, recall, F1-score, and AUC, with values of 95.78%, 85.83%, 84.29%, 83.95% and 90.94%, respectively.
- Subjects
FEATURE selection; LEUKOCYTE count; PULSE wave analysis; BOOSTING algorithms; ATHEROSCLEROSIS; ALGORITHMS; BRACHIOCEPHALIC trunk
- Publication
Journal of Supercomputing, 2024, Vol 80, Issue 9, p13187
- ISSN
0920-8542
- Publication type
Article
- DOI
10.1007/s11227-024-05951-y