Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 2
Encontrada(s) 1 página(s)

RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

Modelos de Machine Learning y estadística multivariante para predecir la posición de los equipos de primera división, Machine Learning and Multivariate Statistics models to predict the position of the first division teams

  • Malagón-Selma, María del Pilar
  • Debón, A.|||0000-0002-5116-289X
  • Ferrer, Alberto|||0000-0001-7244-5947
[EN] This research aims to find which models of Machine Learning and Multivariate Statistics have a greater predictive capacity when deciding what the team's classification will be at the end of the season. The teams that competed in the first division of the Bundesliga, Premier League, LaLiga, Ligue 1 and Serie A throughout the 2018-2019 season have been studied. The badly classified teams by the best of the models, the Random Forest with balanced data, were analyzed in-depth to determine the game's actions that caused the classification error. The results indicate that, generally, the effectiveness in front of goal and the possession of the ball are the statistics in which badly classified teams differ the most with the average of their real position. In conclusion, this research shows how Machine Learning and Multivariate Statistical techniques can be used successfully to discriminate between Top and Bottom teams competing in the best leagues in the world, [ES] Esta investigación tiene como objetivo encontrar qué modelos de Machine Learning y Estadística Multivariante tienen una mayor capacidad de predicción de la posición de los equipos al final de la temporada en la tabla clasificatoria. Se han utilizado las acciones de juego de los equipos que compitieron en la primera división de la Bundesliga, Premier League, LaLiga, Ligue 1 y Serie A, a lo largo de la temporada 2018-2019. Los equipos mal clasificados por el mejor de los modelos, el Random Forest con datos equilibrados, fueron analizados en profundidad para determinar las acciones del juego que provocaban el error de clasificación. Los resultados indican que, generalmente, la efectividad de cara a portería y la posesión del balón son las variables en las que más difieren los equipos mal clasificados respecto a los valores medios de las variables en su grupo real. En conclusión, esta investigación muestra cómo las técnicas de Machine Learning y Estadística Multivariante se pueden utilizar con éxito para predecir la clasificación final de los equipos que compiten en las mejores ligas del mundo., Los autores quieren agradecer a la Universitat Politècnica de València el apoyo económico a través de la beca FPI-UPV (PAID-01-19). Además, agradecen el trabajo de los dos revisores anónimos y el editor, cuyas sugerencias mejoraron el manuscrito original.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publicación Artículo científico (JournalArticle)

Measuring the popularity of football players with Google Trends

  • Malagón-Selma, María del Pilar
  • Debón, A.|||0000-0002-5116-289X
  • Domenech, Josep|||0000-0002-7302-5810
[EN] Google Trends is a valuable tool for measuring popularity since it collects a large amount of information related to Google searches. However, Google Trends has been underused by sports analysts. This research proposes a novel method to calculate several popularity indicators for predicting players¿ market value. Google Trends was used to calculate six popularity indicators by requesting information about two football players simultaneously and creating popularity layers to compare players of unequal popularity. In addition, as the main idea is to obtain the popularity indicators of all players on the same scale, a cumulative conversion factor was used to rescale these indicators. The results show that the proposed popularity indicators are essential to predicting a player¿s market value. In addition, using the proposed popularity indicators decreases the transfer fee prediction error for three different models that are fitted to the data using the multiple linear regression, random forest, and gradient boosting machine methods. The popularity indicator Min, which is a robust reflection of the popularity that represents a player¿s popularity during the periods when they are less popular, is the most important popularity indicator, with a significant effect on the market value. This research provides practical guidance for developing and incorporating the proposed indicators, which could be applied in sports analytics and in any study in which popularity is relevant., P.M-S. Economic support through the FPI-UPV scholarship (PAID-01-19) to the Universitat Politecnica de Valencia. http://www.upv.es/entidades/VINV/indexc.html P.M-S, A.D., and J.D. Grant PID2019-107765RB-I00, funded by MCIN/AEI/10.13039/501100011033. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.