Leveraging Heart Rate Variability and Machine Learning for Daily Activity Classification in Healthy Adult Males

Advancements in machine learning techniques for analyzing heart rate variability (HRV) show great potential in accurately measuring various activities in the field of health monitoring. This work examines the feasibility of using non-invasive HRV data gathered during everyday activities to train machine learning models for accurate activity classification. In the field of personalization, the study focuses on the difficulty of adapting to individual differences in baseline HRV. It also acknowledges that HRV is influenced by stress, physical activity, and external influences, and that it is a dynamic measure. The experimental methodology consists of a pilot study using a retrospective observational cohort approach. It utilizes a waistband device to continuously record heart rate variability (HRV) during different daily activities. The study utilizes a thorough methodology that includes statistical data description, correlation analysis, variance analysis, outlier detection, and identification of cycle periods using power spectral density. The performance of various machine learning models, such as Random Forest, Naïve Bayes, Decision Tree, and Logistic Regression, was evaluated for activity classification. The Decision Tree model exhibited superior performance. The work provides useful insights into the potential of machine learning for HRV-based activity monitoring, highlighting the importance of collecting reliable data and developing tailored models to improve accuracy and practicality.