Objective To find the relationship between physical activities (PA) and HR change by building an effective PA-based HR prediction model.Methods Six models were built according to different PA signal analysis scheme (preprocessing), different NN structures and training algorithms, and different prediction steps (single-step or multi-step).Then the models were tested by using the data collected from real life and comparisons were made between the results.Results The average prediction errors of the models were restricted in a small range.Conclusions The experimental results demonstrated that models built by NN could effectively reflect how PA affect HR, and the comparison results illustrated that the model built by neuro-evolution of augmenting topologies (NEAT) got the best performance in single-step prediction.And Adams-Bashforth technique was the best choice in multi-step prediction.
|