基于混合变换器-贝叶斯框架的微睡眠心血管风险预测模型

Cardiovascular disease (CVD) remains the leading cause of death globally, making early detection of asymptomatic risk factors crucial for preventive interventions. Conventional risk scores, such as Framingham and ASCVD, rely on static risk factors and fail to capture dynamic physiological changes influenced by lifestyle and sleep patterns. Recently, lifelog data collected from smartphones, wearables, and ambient sensors have provided granular, temporally resolved behavioral signals. However, existing lifelog-based models either ignore inter-modal dependencies or treat uncertainty deterministically, leading to overconfident predictions. Additionally, feature selection is often fixed, neglecting user adaptation. This research proposes a hybrid Transformer-Bayesian framework that leverages lifelog data for cardiovascular risk prediction, enhancing accuracy and reliability by accounting for dynamic features and user adaptability.