Abstract
Cardiovascular disease (CVD) is one of the leading causes of death worldwide, early and accurate prediction is crucial for reducing both incidence and mortality rates. Aiming at the problems of insufficient feature association mining and poor generalisation ability of statically weighted ensemble in existing models, this paper proposes a CVD prediction method based on the ensemble of cross-combination strategy and dynamic weighted stacking ensemble (CCS-DWSE). The method firstly constructs a heterogeneous base model through a full combination of multi-feature selection techniques and multi-paradigm classifiers. Next, it designs a dynamic weighted stacking ensemble framework that adjusts the base model weights in real time, and adaptively fuses the prediction results of the K-Nearest Neighbors (KNN) meta-model. Finally, SHAP-based interpretability analysis is employed to quantify the contribution of each feature. The experimental results show that CCS-DWSE achieves accuracies of 98.32%, 92.33%, and 82.10% on three public datasets, with an area under the curve (AUC) of up to 0.9992, representing a 2.76% improvement over the best existing models. This study provides an efficient and scalable solution for CVD prediction and offers a novel perspective for collaborative multi-model learning on complex medical data.