Modeling of ambulatory heart rate using linear and neural network approaches

This chapter presents new results on modeling 24 hour (circadian) human heart rate data collected with the LifeShirt system using a variety of linear regression and neural network models. Such modeling is important in biopsychology, chronobiology, and chronomedicine where signals collected continuously from human subjects for one or several days need to be interpreted. Ambulatory heart rate is influenced by a variety of factors, including physical activity, posture, and respiration, and our models try to predict heart rate based on these factors. The analyses described in the chapter indicate that neural and especially neuro-fuzzy techniques provide better results in the modelling of human heart rate at the circadian scale than conventional linear regression. The advantages of the neuro-fuzzy approaches consist in their computational efficiency, better interpretability, and the possibility of incorporation of prior knowledge for easier model construction.