Scholarlink

<p>Accurate and continuous monitoring of sweat electrolyte concentrations is critical for managing hydration status, thermoregulation, and electrolyte balance, especially in athletic, occupational, and clinical settings. While recent advances in wearable sweat sensors have enabled real-time data acquisition, predictive modeling of sweat analytes remains limited by simplistic estimation techniques and insufficient integration of physiological context. In this study, we propose a data-driven approach to predict concentrations of sodium (Na⁺), potassium (K⁺), and chloride (Cl⁻) in sweat using neural networks trained on a synthetically generated dataset designed to replicate realistic physiological conditions. We evaluated three neural network models of increasing sophistication: a baseline model trained on raw features, an enhanced model with engineered polynomial features and standardized inputs, and a final model optimized through hyperparameter tuning. Input variables included skin temperature, sweat rate, heart rate, hydration level, ambient humidity, and sensor signal strength. Results demonstrated significant performance gains across modeling stages, with the final tuned neural network achieving <i>R</i><sup>2</sup> scores of 0.878 for sodium, 0.744 for potassium, and 0.801 for chloride. These outcomes substantially outperformed traditional linear regression and random forest models. This work highlights the importance of combining physiologically informed feature engineering with modern deep learning techniques and offers a scalable framework for developing intelligent sweat-based monitoring systems capable of real-time, non-invasive electrolyte prediction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrating Wearable Sensor Data and Neural Networks for Real-Time Electrolyte Prediction

  • 期刊论文

  • 发表日期：

    2025年10月23日

查看全文
  • Dheeraj Kumar

摘要

Accurate and continuous monitoring of sweat electrolyte concentrations is critical for managing hydration status, thermoregulation, and electrolyte balance, especially in athletic, occupational, and clinical settings. While recent advances in wearable sweat sensors have enabled real-time data acquisition, predictive modeling of sweat analytes remains limited by simplistic estimation techniques and insufficient integration of physiological context. In this study, we propose a data-driven approach to predict concentrations of sodium (Na⁺), potassium (K⁺), and chloride (Cl⁻) in sweat using neural networks trained on a synthetically generated dataset designed to replicate realistic physiological conditions. We evaluated three neural network models of increasing sophistication: a baseline model trained on raw features, an enhanced model with engineered polynomial features and standardized inputs, and a final model optimized through hyperparameter tuning. Input variables included skin temperature, sweat rate, heart rate, hydration level, ambient humidity, and sensor signal strength. Results demonstrated significant performance gains across modeling stages, with the final tuned neural network achieving R2 scores of 0.878 for sodium, 0.744 for potassium, and 0.801 for chloride. These outcomes substantially outperformed traditional linear regression and random forest models. This work highlights the importance of combining physiologically informed feature engineering with modern deep learning techniques and offers a scalable framework for developing intelligent sweat-based monitoring systems capable of real-time, non-invasive electrolyte prediction.