Optimizing support vector machines and autoregressive integrated moving average methods for heart rate variability data correction

Heart rate variability (HRV) is the variation in time between successive heartbeats and can be used as an indirect measure of autonomic nervous system (ANS) activity. During physical exercise, movement of the measuring device can cause artifacts in the HRV data, severely affecting the analysis of the HRV data. Current methods used for data artifact correction perform insufficiently when HRV is measured during exercise. In this paper we propose the use of autoregressive integrated moving average (ARIMA) and support vector regression (SVR) for HRV data artifact correction. Since both methods are only trained on previous data points, they can be applied not only for correction (i.e., gap filling), but also prediction (i.e., forecasting future values). Our paper describes: • why HRV is difficult to predict and why ARIMA and SVR might be valuable options. • finding the best hyperparameters for using ARIMA and SVR to correct HRV data, including which criterion to use for choosing the best model. • which correction method should be used given the data at hand.