Wearable‐Based Personalized Exercise Heart Rate Estimation and Distribution Analysis Using Dual‐Context LSTM Model

This study introduces a novel dual‐context Long Short‐Term Memory (LSTM) model for personalized exercise heart rate estimation and distribution analysis based on wearable devices, addressing the limitations of existing wearable‐based methods in capturing the complex, user‐specific heart rate dynamics. Leveraging a large‐scale dataset with over 250 000 workout records, our model incorporates both immediate exercise context and historical user data through a multi‐task learning framework. The architecture features personalized embedding layers accounting for static user attributes and dynamic exercise history, significantly enhancing individualized estimations. Experimental results show that the model outperforms baseline methods, achieving a mean absolute error (MAE) of 5.2 bpm and an R 2 score of 0.89, with robust predictions across various exercise types. Our analysis reveals insightful patterns in how individual characteristics and exercise history influence heart rate responses, contributing to a deeper understanding of personalized exercise physiology. These findings have important implications for developing more effective personalized fitness recommendations and health monitoring systems, advancing the fields of exercise science and personalized health technology.