Análisis retrospectivo del rendimiento de un equipo de futbol profesional costarricense a partir de variables recolectadas por medio de tecnología inercial
Archivos
Fecha
2023
Tipo
tesis de maestría
Autores
Morera Siércovich, Pier Luigi
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
Propósito: El propósito del estudio fue identificar las variables que aumentan las posibilidades de ganar un partido y las variables que explican el rendimiento de futbolistas de un equipo de fútbol profesional costarricense, según la posición en el terreno de juego. Metodología: Se construyó un archivo de 1037 casos de jugadores con datos recopilados durante dos temporadas y se realizó un análisis con algoritmos de machine learning a partir del análisis de datos recolectados con registros semiautomáticos y tecnología inercial. Resultados: Para explicar el rendimiento de los futbolistas se construyeron modelos de regresión lineal múltiple que tuvieron valores predictivos de 0.67 para defensas centrales, 0.56 para defensas laterales, 0.71 para mediocampistas defensivos, 0.82 para extremos, de 0.80 para mediocampistas creativos, y 0.82 para centro delanteros. Las variables con mayor peso en la estimación fueron de naturaleza técnico-tácticas: goles y asistencias (defensas centrales, delanteros extremos y centro delanteros), tiros a portería (defensas centrales, mediocampistas creativos, y centro delanteros), disputas defensivas ganadas (defensas centrales y laterales), porcentaje de efectividad de los pases y disputas por arriba ganadas (laterales, mediocampistas ofensivos y extremos delanteros). En cuanto a la predicción de resultado, el método de bosques aleatorios permitió obtener mayor precisión en la estimación de la clasificación (AUC > 0.5) comparado con árboles de decisión. Las variables que explican mejor cuando se ganan los partidos fueron la máxima velocidad (todas las posiciones excepto los defensas centrales), el número de sprints (defensas centrales, laterales, mediocampistas ofensivos y centro delanteros), pases (defensa central y mediocampista ofensivo), la distancia recorrida a alta intensidad (defensas centrales y mediocampistas ofensivos), y la cantidad de metros por minuto recorridos (mediocampista defensivos, extremo delanteros, y centro delanteros). Conclusión: Los modelos estudiados fueron capaces de asignar una calificación a los jugadores según su posición en el campo e identificar las variables más asociadas con ganar los juegos.
Purpose: The purpose of the study was to identify the variables that increase the chances of winning a game and the variables that explain the performance of soccer players from a Costa Rican professional soccer team, according to the position on the field. Methods: A file of 1037 player cases with data collected during two seasons was analyzed with machine learning algorithms based on data collected with semi-automatic records and inertial technology. Results: Multiple linear regression models explained the performance of the soccer players, which showed predictive values of 0.67 for central defenders, 0.56 for wing defenders, 0.71 for defensive midfielders, 0.82 for wingers, 0.80 for creative midfielders, and 0.82 for center forwards. The variables with the most significant weight in the estimation were technical-tactical: goals and assists (central defenders, extreme forwards, and center forwards), shots on goal (central defenders, creative midfielders, and center forwards), defensive disputes won (central defenders and full-backs), percentage of pass effectiveness and disputes over the top won (full-backs, attacking midfielders and forwards). Regarding the prediction of the result, the random forest method allowed for obtaining greater precision in the estimation of the classification (AUC > 0.5) compared to decision trees. The variables that best explain won games were maximum speed (all positions except central defenders), the number of sprints (central defenders, wings, attacking midfielders, and center forwards), passes (central defender and attacking midfielder), the distance covered at a high intensity (central defenders and attacking midfielders), and the number of meters per minute covered (defensive midfielders, wingers, and center forwards). Conclusion: The models studied assigned a rating to the players according to their position on the field and identified the variables most associated with winning games.
Purpose: The purpose of the study was to identify the variables that increase the chances of winning a game and the variables that explain the performance of soccer players from a Costa Rican professional soccer team, according to the position on the field. Methods: A file of 1037 player cases with data collected during two seasons was analyzed with machine learning algorithms based on data collected with semi-automatic records and inertial technology. Results: Multiple linear regression models explained the performance of the soccer players, which showed predictive values of 0.67 for central defenders, 0.56 for wing defenders, 0.71 for defensive midfielders, 0.82 for wingers, 0.80 for creative midfielders, and 0.82 for center forwards. The variables with the most significant weight in the estimation were technical-tactical: goals and assists (central defenders, extreme forwards, and center forwards), shots on goal (central defenders, creative midfielders, and center forwards), defensive disputes won (central defenders and full-backs), percentage of pass effectiveness and disputes over the top won (full-backs, attacking midfielders and forwards). Regarding the prediction of the result, the random forest method allowed for obtaining greater precision in the estimation of the classification (AUC > 0.5) compared to decision trees. The variables that best explain won games were maximum speed (all positions except central defenders), the number of sprints (central defenders, wings, attacking midfielders, and center forwards), passes (central defender and attacking midfielder), the distance covered at a high intensity (central defenders and attacking midfielders), and the number of meters per minute covered (defensive midfielders, wingers, and center forwards). Conclusion: The models studied assigned a rating to the players according to their position on the field and identified the variables most associated with winning games.
Descripción
Palabras clave
RENDIMIENTO FÍSICO, DEPORTE, MONITOREO, POSICIÓN DE JUEGO, ANÁLISIS DE PARTIDOS, TECNOLOGÍA INERCIAL