Farrahi Vahid, Rostami Mehrdad
Institute for Sport and Sport Science, TU Dortmund University, Dortmund, Germany.
Centre of Machine Vision and Signal Analysis, Faculty of Information Technology, University of Oulu, Oulu, Finland.
J Act Sedentary Sleep Behav. 2024 Jan 30;3(1):5. doi: 10.1186/s44167-024-00045-9.
The nature of human movement and non-movement behaviors is complex and multifaceted, making their study complicated and challenging. Thanks to the availability of wearable activity monitors, we can now monitor the full spectrum of physical activity, sedentary, and sleep behaviors better than ever before-whether the subjects are elite athletes, children, adults, or individuals with pre-existing medical conditions. The increasing volume of generated data, combined with the inherent complexities of human movement and non-movement behaviors, necessitates the development of new data analysis methods for the research of physical activity, sedentary, and sleep behaviors. The characteristics of machine learning (ML) methods, including their ability to deal with complicated data, make them suitable for such analysis and thus can be an alternative tool to deal with data of this nature. ML can potentially be an excellent tool for solving many traditional problems related to the research of physical activity, sedentary, and sleep behaviors such as activity recognition, posture detection, profile analysis, and correlates research. However, despite this potential, ML has not yet been widely utilized for analyzing and studying these behaviors. In this review, we aim to introduce experts in physical activity, sedentary behavior, and sleep research-individuals who may possess limited familiarity with ML-to the potential applications of these techniques for analyzing their data. We begin by explaining the underlying principles of the ML modeling pipeline, highlighting the challenges and issues that need to be considered when applying ML. We then present the types of ML: supervised and unsupervised learning, and introduce a few ML algorithms frequently used in supervised and unsupervised learning. Finally, we highlight three research areas where ML methodologies have already been used in physical activity, sedentary behavior, and sleep behavior research, emphasizing their successes and challenges. This paper serves as a resource for ML in physical activity, sedentary, and sleep behavior research, offering guidance and resources to facilitate its utilization.
人类运动和非运动行为的本质复杂且多面，这使得对其进行研究既复杂又具有挑战性。得益于可穿戴活动监测器的出现，我们现在能够比以往任何时候都更好地监测身体活动、久坐行为和睡眠行为的全貌——无论受试者是精英运动员、儿童、成年人还是已有疾病的个体。生成的数据量不断增加，再加上人类运动和非运动行为固有的复杂性，因此需要开发新的数据分析方法来研究身体活动、久坐行为和睡眠行为。机器学习（ML）方法的特性，包括其处理复杂数据的能力，使其适用于此类分析，从而可以成为处理此类性质数据的替代工具。ML有可能成为解决许多与身体活动、久坐行为和睡眠行为研究相关的传统问题的出色工具，例如活动识别、姿势检测、特征分析和相关性研究。然而，尽管有这种潜力，ML尚未被广泛用于分析和研究这些行为。在本综述中，我们旨在向身体活动、久坐行为和睡眠研究领域的专家——那些可能对ML不太熟悉的人——介绍这些技术在分析他们的数据方面的潜在应用。我们首先解释ML建模流程的基本原理，强调应用ML时需要考虑的挑战和问题。然后我们介绍ML的类型：监督学习和无监督学习，并介绍一些在监督学习和无监督学习中常用的ML算法。最后，我们重点介绍ML方法已用于身体活动、久坐行为和睡眠行为研究的三个领域，强调它们的成功和挑战。本文作为身体活动、久坐行为和睡眠行为研究中ML的资源，提供指导和资源以促进其应用。
