Project Background: Sporting almanacs have long collected statistical data from football league
seasons and analysts have sought to use this data to predict the winners of future games. In
recent decades, companies and fans have collected vast amounts of data from games seeking to
predict game outcomes and understand the metrics responsible for a game win. Due to their
low number, goals scored are not in themselves an appropriate metric for analysis. Instead,
in-game tracking data (obtained through computer vision, machine learning and the \Internet
of Things") promises to provide greater insight into soccer matches that can be strategically
leveraged by teams to increase the probability of a win.
Aims/Objectives: This project shall work with The Faculty of Sports and Exercise to apply
different Data Science techniques to existing soccer related data sets. Specifcally, the focus is
to reveal dynamic grouping patterns in phases of a match using clustering techniques. In this
project the student shall employ one or two clustering techniques to dynamic soccer datasets
to determine whether greater game insight can be achieved through their use. The outputs
of these clustering techniques shall be compared against information already achieved from
existing research to assess their effectivenss. Furthermore, the project seeks to understand why
these clustering techniques work or don't work in whole or in part and what alterations could
be made to the algorithms to increase their effectiveness.
Methodological approach: two clustering techniques
Dataset(s): Positional data (x, y coordinates) of twenty-two players made available by The
Faculty of Sports and Exercise.
Resources required: Own computer, coding using Python or R.