Exploring body composition and physical condition profiles in relation to playing time in professional soccer: a Principal Components Analysis and Gradient Boosting approach

David Ulloa-Díaz^1*Carlos Gabriel Fàbrica^1,2Carlos Jorquera³Francisco Guede- Rojas⁴Jorge Pérez-Contreras⁵Demetrio Lozano Jarque⁶Claudio Carvajal⁷Luis Romero-Vera⁸

• ¹Department of Sports Sciences and Physical Conditioning, Universidad Catolica de la Santisima Concepcion, Concepción, Chile
• ²Unidad Académica de Biofísica, Facultad de Medicina, Universidad de la Republica Uruguay, Montevideo, Uruguay
• ³Facultad de Ciencias, Escuela de Nutrición y Dietética, Universidad Mayor, Santiago, Chile
• ⁴Exercise and Rehabilitation Sciences Institute, School of Physical Therapy, Faculty of Rehabilitation Sciences, Universidad Andres Bello, Santiago, Chile
• ⁵Escuela de Ciencias del Deporte y Actividad Física, Facultad de Salud, Universidad Santo Tomas, Santiago, Chile
• ⁶Valora Research Group, Health Sciences Faculty, Universidad San Jorge, Villanueva de Gállego, Spain
• ⁷Facultad de Ciencias de la rehabilitación y Calidad de vida, Escuela de kinesiología, Universidad San Sebastian, Concepción, Chile
• ⁸Facultad de Salud y Ciencias Sociales, Escuela de Ciencias de la Actividad Física, Universidad de Las Americas, Santiago, Chile

Background: This study aimed to explore whether a predictive model based on body composition and physical condition could estimate seasonal playing time in professional soccer players. Methods: 24 professional soccer players with 5 to 7 years of professional experience participated. Body composition and physical condition variables were assessed, and total minutes played during the season were recorded as the dependent variable. Correlations between variables were examined to reduce multicollinearity, followed by a principal component analysis (PCA) of the selected predictors. The first three components were used as inputs in a Gradient Boosting model. Model performance was evaluated using 5-fold cross-validation and leave-one-out cross-validation (LOOCV). Results: High intercorrelations among independent variables (r > 0.70) justified dimensionality reduction through PCA. The first three components explained 70% of the total variance. However, no direct correlations were observed between individual variables and minutes played, and the Gradient Boosting model did not achieve positive predictive performance under cross-validation (5-fold CV: R² = −0.04; LOOCV: R² < 0). Conclusion: In this small dataset, a multivariate approach combining PCA and Gradient Boosting did not yield predictive accuracy for playing time. Nonetheless, the PCA revealed meaningful structures in the players' physical and body composition profiles, which may inform future research. Larger and more heterogeneous samples are required to determine whether component-based predictors can reliably estimate playing time in professional soccer.