Body Composition Profiles of Applicants to a Physical Education and Sports Major in Southeastern Mexico

Abstract

This study aimed to determine the body composition profile of candidates applying for a Physical Education and Sports major. 327 young adults (F: 87, M: 240) participated in this cross-sectional study. Nutritional status and body composition analysis were performed, and the profiles were generated using an unsupervised machine learning algorithm. Body mass index (BMI), percentage of fat mass (%FM), percentage of muscle mass (%MM), metabolic age (MA), basal metabolic rate (BMR), and visceral fat level (VFL) were used as input variables. BMI values were normal-weight although VFL was significantly higher in men (<0.001; η2 = 0.104). MA was positively correlated with BMR (0.81 [0.77, 0.85]; p < 0.01), BMI (0.87 [0.84, 0.90]; p < 0.01), and VFL (0.77 [0.72, 0.81]; p < 0.01). The hierarchical clustering analysis revealed two significantly different age-independent profiles: Cluster 1 (n = 265), applicants of both sexes that were shorter, lighter, with lower adiposity and higher lean mass; and, Cluster 2 (n = 62), a group of overweight male applicants with higher VFL, taller, with lower %MM and estimated energy expended at rest. We identified two profiles that might help universities, counselors and teachers/lecturers to identify applicants in which is necessary to increase physical activity levels and improve dietary habits.

Autoconcepto físico en estudiantes universitarios: Generación de perfiles por clasificación jerárquica sobre componentes principales (Physical self-concept in university students: Generating profiles with hierarchical classification on principal components)

“…No obstante, se observó una mayor proporción de mujeres en el perfil con menor autoconcepto físico. El hecho que el sexo tenga alta relevancia en la proporción de sujetos en los perfiles también ha sido demostrado previamente por nuestro grupo de investigación en cuanto a la composición corporal (Gasperin-Rodriguez et al, 2022) y la aptitud física (Bonilla, Sanchez-Rojas, et al, 2022) en estudiantes universitarios latinoamericanos. De manera interesante, los estudiantes de Psicología representan la mayor proporción (75.89%) de participantes agrupados en el Perfil 2.…”

“…La mayor parte de la varianza de los datos es explicada por los componentes principales y el número de componentes nos indica la existencia o no de diversidad en el tipo de patrones. Como hemos desarrollado anteriormente (Bonilla, Peralta-Alzate, et al, 2022;Bonilla, Sanchez-Rojas, et al, 2022;Cardozo et al, 2021;Gasperin-Rodriguez et al, 2022), los participantes se subdividieron en clusters mediante aprendizaje automático no supervisado para identificar puntos de datos similares (agrupaciones naturales) y extraer los patrones de perfil. Para esto se ejecutó un análisis de Agrupación Jerárquica en Componentes Principales utilizando la función HCPC del paquete 'FactoMi-neR' en el entorno de programación R (R Statistics, versión 4.0.5, Inc, Colorado, USA).…”

Profiling Physical Fitness of Physical Education Majors Using Unsupervised Machine Learning

“…Eta-squared (η 2 ) was used to report the magnitude of differences assuming 0.09, 0.14, and >0.22 as a small, medium, and large effect size [28]. As we have performed previously [29][30][31][32], the participants were subdivided into clusters using unsupervised machine learning to identify similar data points (natural groupings) and extract the profile patterns. We used the partitioning around the medoids (PAM) algorithm, also known as k-Medoids clustering which, unlike the k-means algorithm, considers the median as the center of a cluster, thus, it is more robust to noises and outliers [33].…”

Artificial Intelligence Applications in Digital Anthropometry ^{<a class="tippyShow" data-tippy-arrow="true"data-tippy-content="Presented as a keynote at the 'I Congreso Iberoamericano de Antropometría Aplicada' organized by the 'Red Iberoamericana de Investigadores en Antropometría Aplicada'." data-tippy-interactive="true"data-tippy-theme="light-border"style="cursor:pointer">1</a>}

“…For example, unsupervised ML methods allow the inherent structure of unlabeled data to be discovered. Since the clustering method is applied to find homogeneous subgroups of observations, the algorithms have been used to profile phenotypes in sports science [3][4][5]. It is worth mentioning that the selection of the most appropriate clustering algorithm will depend on the distribution of the data and the phenotype/biological phenomenon to be analyzed (field knowledge) [6].…”

“…After a possible transition into digital anthropometry, basic measures (stature, sitting height, and arm span), girths, breadths, and lengths can be evaluated with high accuracy and reliability using AI methods. Finally, unsupervised ML algorithms would benefit sports practitioners when the aim is to obtain anthropometric profiles, as has been demonstrated by our research group [3][4][5].…”