Optimization of Deep Learning Models for Enhanced Respiratory Signal Estimation Using Wearable Sensors

Measuring breathing changes during exercise is crucial for healthcare applications. This study used wearable capacitive sensors to capture abdominal motion and extract breathing patterns. Data preprocessing methods included filtering and normalization, followed by feature extraction for classification. Despite the growing interest in respiratory monitoring, research on a deep learning-based analysis of breathing data remains limited. To address this research gap, we optimized CNN and ResNet through systematic hyperparameter tuning, enhancing classification accuracy and robustness. The optimized ResNet outperformed the CNN in accuracy (0.96 vs. 0.87) and precision for Class 4 (0.8 vs. 0.6), demonstrating its capability to capture complex breathing patterns. These findings highlight the importance of hyperparameter optimization in respiratory monitoring and suggest ResNet as a promising tool for real-time assessment in medical applications.