Heart Rate Modeling and Prediction Using Autoregressive Models and Deep Learning

Area of Science:

• Physiological time series analysis
• Biomedical signal processing
• Machine learning in healthcare

Background:

• Physiological time series, like heart rate, are complex, nonlinear, and nonstationary.
• Heart rate variability is crucial for identifying cardiovascular, respiratory, and mood disorders.
• Accurate heart rate prediction is vital for disease prevention and management.

Purpose of the Study:

• To compare the predictive performance of three forecasting models: Autoregressive Model (AR), Long Short-Term Memory Network (LSTM), and Convolutional Long Short-Term Memory Network (CLSTM).
• To identify the best-performing model for modeling and forecasting heart rate behavior.
• To evaluate model accuracy using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).

Main Methods:

• Collected 10-day minute-by-minute heart rate data from twelve diverse participants using a wearable device.
• Trained and tested AR, LSTM, and CLSTM models on the collected heart rate data.
• Quantified model performance using MAE and RMSE metrics.

Main Results:

• All three models demonstrated comparable performance, but the Autoregressive Model consistently yielded the best results across all tested scenarios.
• For instance, the AR model achieved a lower MAE (2.069) compared to LSTM (2.173) and CLSTM (2.138) for one participant.
• The study observed that heart rate is largely dependent on recent patterns, supporting its treatment as an autoregressive process.

Conclusions:

• The Autoregressive Model provides accurate minute-by-minute heart rate prediction, even in individuals with varied demographics and lifestyles.
• Linear models are effective for heart rate forecasting in healthy individuals, suggesting simplicity can be advantageous.
• The AR model's efficacy opens avenues for applications in arrhythmia detection and training response analysis.