Print Email Facebook Twitter Classification of playing styles in football Title Classification of playing styles in football: The use of ball action data Author Wensveen, C.J. Contributor Jongbloed, G. (mentor) Faculty Electrical Engineering, Mathematics and Computer Science Department Delft Institute of Applied Mathematics Programme Probability, Risk and Statistics Date 2016-10-05 Abstract Coaches can benefit from objective information about playing styles applied in football matches. In this work, two methods for determining (characteristics of) the playing style of a football team in a certain match based on statistics of ball actions are constructed. The first method assigns a match to one of four commonly applied playing styles in football based on a set of benchmark matches. Within this method, a relevant variable set with respect to the playing style of a team is selected based on the so-called minimum-redundancy-maximum-relevance algorithm. This algorithm makes use of mutual information as a measure of relevance. The mutual information between variables is estimated by the so-called Kraskov and adjusted Kraskov estimator. After a relevant variable set has been found, matches are assigned to one of the four playing styles by the use of a combination of a hierarchical scheme of K-means clustering and 1-Nearest Neighbors. The second method focuses on general playing style characteristics of matches as opposed to labeling a match with a specific playing style. This way, details about the playing style in a match can be obtained without limiting to the four prior labeled playing styles. Using principal component analysis combined with domain knowledge, three characteristic variables are created which together give a general overview regarding the playing style applied in a match. Application of both models on different sets of matches show satisfying results which agree with domain know-ledge. These models can be used to provide football coaches with information regarding playing styles applied in matches. Coaches can use this information in order to both evaluate their own team as well as analyze their opponents. To reference this document use: http://resolver.tudelft.nl/uuid:9040d3df-ae22-4537-82b2-4d084dd5b01c Part of collection Student theses Document type master thesis Rights (c) 2016 Wensveen, C.J. Files PDF Thesis-confidential version.pdf 1.14 MB Close viewer /islandora/object/uuid:9040d3df-ae22-4537-82b2-4d084dd5b01c/datastream/OBJ/view