Digital health interventions from the humanistic perspective of sports: strategies to promote health for all.

Introduction: Personalized exercise recommendations play a critical role in promoting sustainable health. Traditional models often fail to account for individual demographic and physiological variability. A humanistic deep learning framework can bridge this gap by adapting health strategies to diverse needs. Methods: We propose a deep learning architecture that integrates normalized exercise features such as Age, BMI, Gender, Exercise Type, Heart Rate, and Duration. The model employs feature embedding, residual connections, and multi-head attention mechanisms to dynamically prioritize physiologically important features. Results and discussion: The proposed model achieved a 35% relative reduction in Mean Absolute Error (MAE) compared to competitive machine learning and deep learning baselines. In classification tasks, it improved Accuracy and F1-score by 9-12%, reaching an Accuracy of 95.1% and an F1-score of 94.7%. The proposed framework establishes a new benchmark for personalized digital health interventions by combining predictive accuracy and humanistic fairness. It demonstrates the feasibility of delivering individualized exercise strategies at scale, while ensuring equitable health promotion across diverse populations.