Predicting pulse wave age from cardiovascular characteristics using machine learning algorithms

Abstract

Arterial pulse waves are an essential and informative source of data for measuring cardiovascular health and are currently available on various devices, such as smartwatches. In addition, most experiments have focused on pulse wave velocity (PWV) and arterial pulse waveform (APW). However, in recent years, due to many factors, such as high work pressure among young people, cardiovascular diseases have become more prevalent. As a result, we need more research on pulse age to help patients focus on the consistency of their biological age with their legal age. Therefore, this paper introduces a machine learning framework to predict the age of patients according to the HR and SV data in the pulse wave of existing patients. At the beginning of the experiment, we attempted to use K-nearest neighbours (KNN), logistic regression (LR), and random forest (RF) models. Then we compared the accuracy of patient age estimates; RF had the best performance and was used as the final model. In addition, HR and SV were selected as the main features to predict patients' age according to the feature's importance. The final experimental results indicate that the RF model can predict the results with up to 1 accuracy only with two features, i.e., HR and SV. In particular, this is the point that we need to continue to pay attention to in the future to determine whether the prediction accuracy of this model can reach 1. Is it because the model is overly optimistic or because the selected HR and SV features strongly correlate with age? In the future, we can introduce the concept of actual body age and attract young people's attention to their body age. For example, more people will be urged to take a quick and efficient pulse diagnosis, complete a simple physical examination, and give rapid feedback by placing devices in places where young people gather, such as shopping malls. In this case, users can immediately know what part of their body may have problems. In general, our experiment demonstrates that patients' pulse wave ages could be well predicted based on their HR and SV data with high accuracy through the ML model of RF.