Abstract
One of the most important indicators of physiological health is the heart rate (HR), commonly used to understand exercise intensity, maximize training, and prevent overtraining or injury during endurance running. HR prediction accuracy is the key to building training strategies that are personalized wearable technology when influenced by noise or measurement errors. This paper review presents an analysis of the current ways of predicting heart rate during running, using modern machine learning or deep learning algorithms and the traditional statistical techniques. Conventional paradigms like linear regression are interpretable but usually restricted to nonlinear physiological responses. Machine learning algorithms, such as support vector machines, decision trees, and ensemble models have shown higher accuracy through the combination of several sensor-based variables, such as pace, distance, and workload. In more modern times, deep learning structures, especially recurrent and convolutional neural networks, have demonstrated a good promise in the modelling of complex temporal relationships in HR data. With these developments, issues persist in the fields of variability of data, model generalization and interpretability. This review has identified the present success, the major shortcomings, and future opportunities in the development of individualized, adaptive and explainable HR prediction models to improve performance and health in running.
Description
null