Random vector functional link network based on multi-scale kernel for stroke risk assessment

doi:10.1016/j.bspc.2025.108681

Random vector functional link network based on multi-scale kernel for stroke risk assessment

2025-09-23

0

PRE

AI

Y

Ying Liu

C

C. Madhusudhana Rao *

DOI：10.1016/j.bspc.2025.108681

Original

Original request for help

Save

Abstract

En 中文

Stroke, a leading cause of global disability and mortality, has seen rising incidences at younger ages, escalating its socio-economic impact. Addressing the urgent need for effective prevention and treatment, this study proposes a novel approach, i.e., random vector functional link network with multi-scale kernel (MSK-RVFLN) for stroke risk assessment. Our approach originates from modifying the traditional Radial Basis Function (RBF) with the Large Margin Nearest Neighbor (LMNN) algorithm, resulting in a new Multi-Scale Kernel (MSK). This kernel combines RBF’s local feature sensitivity with the structural of data distribution, employing Density Peaks Clustering (DPC) for data segmentation to optimize kernel scaling. The efficacy of the proposed MSK-RVFLN is demonstrated through a stroke risk prediction dataset. The MSK-RVFLN achieved Accuracy, F1-Score, G-mean, and Area Under Curve (AUC) values of 93.87 %, 93.97 %, 93.86 %, and 97.42 %, respectively, surpassing conventional kernels and outperforming standard machine learning and similarity models. Comprehensive evaluations across multiple datasets confirm the model’s robustness and superior stroke risk prediction capabilities, establishing a solid foundation for medical diagnostics.

Random vector functional link network based on multi-scale kernel for stroke risk assessment

Abstract

Journal

Organization