Deep Learning Approaches to Predict Exercise Exertion Levels Using Wearable Physiological Data.

Area of Science:

• Sports Science
• Biomedical Engineering
• Machine Learning

Background:

• Wearable technology offers novel ways to collect physiological data during exercise.
• Accurate prediction of exercise exertion is crucial for personalized training and health monitoring.
• Existing methods for assessing exertion may be subjective or require specialized equipment.

Purpose of the Study:

• To develop and evaluate deep learning models for predicting exercise exertion levels.
• To utilize physiological data from wearable sensors (ECG, pulse oximeter) for exertion prediction.
• To compare classification and regression models in forecasting perceived exertion.

Main Methods:

• Collected real-time ECG, pulse rate, oxygen saturation, and RPM data during cycling sessions.
• Calculated heart rate variability (HRV) features from ECG data.
• Engineered predictive features by averaging data within 2-minute windows.
• Employed feature selection algorithms to identify optimal predictors.
• Trained and tested deep learning models for regression and classification tasks.

Main Results:

• Deep learning models achieved high performance during training, with accuracy and F1 scores up to 98.2% and 98%, respectively.
• Model testing yielded the highest accuracy and F1 scores of 80%.
• Physiological data, including heart rate and HRV, proved effective in predicting exertion levels.

Conclusions:

• Deep learning models show promise for accurately predicting exercise exertion using wearable sensor data.
• The study demonstrates the feasibility of unobtrusive, real-time exertion monitoring.
• Further research can refine these models for broader applications in fitness and healthcare.