| Summary: | In any team sport the success or failure of that team depends on performances of its
individual players in their respective roles, and the effectiveness of interactions between
players as they work together in the team. Information on the exact demands of a
player���s role and an effective and objective way to evaluate their performance will
provide insight into their tactical strengths and weaknesses.
A system for measuring and evaluating a netball player���s performance on court is
described, and statistical methods such as cluster analysis were used to firstly profile the
different playing positions, and then also to classify players within each position by their
strengths, weaknesses and tactical preferences. These results provide coaches with
another tool to use for selection and strategy decisions.
A comprehensive list of variables was developed in consultation with elite netball
coaches and players and data were sourced from television footage of the 2008 ANZ
Championship netball competition. Each of the seven playing positions was profiled
using simple exploratory analysis and then nine different clustering algorithms were
compared in their ability to correctly classify players into their correct positions. The
same algorithms were then used to identify 3-4 types of player within each position,
according to their tactical differences within the game.
Model-based clustering produced the most accurate partitioning of players into playing
positions, with an accuracy of 85%, followed by the e-distance and Ward���s methods on
78% each. At the other end of the scale was single linkage, with a best accuracy of just
38%. Kernel k-means, which was designed to be an improvement on ordinary k-means,
also performed poorly (59%). In the final stage of the project there were 3-4 player types
successfully identified within each position, providing insight into how different players
approach the game, and how to utilise their different strengths when planning game
strategy.
|
|---|
