Heart disease risk prediction based on deep learning multi-scale convolutional enhanced Swin Transformer model

Heart disease is a leading global cause of death, making early prediction critical. This study proposes a multi-scale convolution-enhanced Swin Transformer (MSCST) model for heart disease risk assessment. The model employs a multi-branch convolutional network with channel attention to extract and optimize multi-scale features. These features are processed by a Swin Transformer module to integrate global and local information via self-attention. SHAP analysis is incorporated to enhance interpretability. Evaluated on the Cleveland Heart Disease dataset, MSCST achieved 89.42% accuracy and an AUC of 0.8908, outperforming both traditional machine learning and existing deep learning methods.