Abstract
Research in field sports often measures the performance of players during competitive games with individual and time-based descriptive statistics. Data is generated using GPS technologies, capturing simple data such as time (seconds) and position (latitude and longitude). While the data capture is highly granular and in relatively high volumes, the raw data are unsuited to any form of analysis or machine learning functions. The dataset presented here is created through a data engineering process, driven by domain experts, to transform the GPS coordinates into a series of (player) actions. Using 14 outfield players from each of 11 games, we present a database comprising 12 variables and almost 160k actions. Its reuse potential is targeted at machine learning researchers, sport scientists and coaches who may have different requirements represented as different analytical queries. This dataset is dimensional in nature, facilitating a rich set of analytics across dimensions such as game, player, action type and duration.