Classification of Goalkeeper Dives: Data Analysis Report

Carroll, Daragh (2023) Classification of Goalkeeper Dives: Data Analysis Report. Undergraduate thesis, Dublin, National College of Ireland.

Preview

PDF (Bachelor of Science) Download (1MB) | Preview

Abstract

Performance analysis plays a major role in helping professional footballers compete at the highest level. Through technological advances such as GPS trackers, data relating to player performance during matches and training sessions can be gathered and clubs now need a way of leveraging this data. One such area relates to the analysis and classification of goalkeeper actions, to create individual training programmes and improve performance.

The goal of this study is to investigate whether machine learning techniques can be a used to assist in classifying goalkeeper actions, firstly as dives/non-dives and then based on the nature of those dive e.g. (Left High Dive).

A dataset built on GPS tracker data and video footage from goalkeeper training sessions, at multiple clubs across Europe and America, together with goalkeeper characteristics and weather data, which had been manually classified provided a rich dataset on which to train and test machine learning models.

A Crisp-DM approach was implemented to train, test and validate a comprehensive set of classification models including, logistic regression, k nearest neighbours, decision trees, support vector classification and random forest. Each model was run iteratively on the dataset with different levels of pre processing applied. Cross validation was used to validate the models.

The study found that the best model for binary classification of goalkeeper actions was the random forest, with minimal pre-processing techniques applied to the dataset (Cross-Val = 0.86, Accuracy = 0.84, Precision = 0.84, Recall = 0.84, F1-score = 0.84). For the subsequent multi-class classification of those dives by type the random forest was the best performing model, with minimal pre-processing techniques applied to the dataset (Cross-Val = 0.81, Accuracy = 0.81, Precision = 0.83, Recall = 0.82, F1-score = 0.81). The results show that machine learning techniques can be successfully implemented on goalkeeper performance data, with high levels of accuracy.

Item Type:	Thesis (Undergraduate)
Supervisors:	Name Email -, - UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science G Geography. Anthropology. Recreation > GV Recreation Leisure > Sports > Soccer Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Bachelor of Science (Honours) in Computing
Depositing User:	Tamara Malone
Date Deposited:	15 Jan 2024 17:48
Last Modified:	15 Jan 2024 17:48
URI:	https://norma.ncirl.ie/id/eprint/6919

Actions (login required)

View Item