Author(s)

Yuang Liu

Corresponding Author

Yuang Liu

ABSTRACT

This paper aims to use the nature of the network and the analysis methods of social networks to analyze the passing behavior of football players in the game, and give a set of quantitative methods to evaluate team performance, and finally hope to select some reasonable game process indicators and predict the outcome of the game through machine learning. The data set used is huskies team’s real game data. We combined the network model and players' position to figure out the dyadic, triadic configurations, and the structure of the team. After that, we use Centrality and Closeness Centralization to evaluate players, use Weighted Directed Network Entropy and Clustering Coefficient to evaluate team. What’s more, we have constructed several indicators from the data to measure the ability of players and the performance that reflect successful teamwork. Firstly, we have established passing networks for huskies, including a diagram of the passing network based on passing data from the entire season, and several passing networks of different matches. From the former, we can see the passing situation of the entire team throughout the season, and in the latter, we can see the changes in the network during the season. We draw a schematic diagram that both reflects the player's actual passing position and reflects the passing network between players. Such a graph is handy for analyzing the dyadic and triadic configurations and team formations. At the same time, we study the tactics of our struggle against the enemy, figuring out structures and critical attacking paths of the Huskies in several matches. This is followed by individual and team evaluations. We defined evaluation indexes and established evaluation models for them. For individuals, we selected several representative members in the Huskies team and analyzed their characteristics through data. As for the team, we analyzed the changes in the evaluation factors over the course of the season. We use BP Neural Network to forecast the matches outcome and find out the critical features of a team to win. We use degree centrality to measure the centralization of the team, use eigenvector centrality to measure the importance of players. For the entire passing network, we use closeness centralization to measure team-centricity and Weighted directed network entropy to measure team decentralized level. Then we find how the centralization of the team impacts the behavior during the match.

KEYWORDS

Team performance, Social network analysis, Competition result prediction