Playing style identification in team sports: A systematic review from statistical dimensionality reduction to unsupervised machine learning

Abstract

Identifying team playing styles is crucial for tactical planning, player recruitment, and performance optimization in team sports. With the increasing availability of match sheet, event, and tracking data, statistical dimensionality reduction and unsupervised machine learning (UML) have become valuable for uncovering hidden patterns in collective behaviours. This systematic review synthesizes evidence from 25 peer-reviewed studies, identified through a PRISMA-based search of Web of Science, PubMed, Scopus, and SPORTDiscus, to evaluate current applications of these techniques in classifying team playing styles. Three main findings emerged. First, principal component analysis (PCA) is the most frequently applied method (44%), typically combined with clustering algorithms (e.g., k-means) to classify teams by playing style and analyse their associations with match outcomes or tactical patterns over a season. Second, research is heavily concentrated on soccer (84%), with limited investigation in other sports such as rugby and basketball, restricting the generalisability of findings. Third, marked differences in feature selection, data preprocessing, and reporting approaches reduce methodological transparency and limit both the reproducibility and comparability of results. Despite these limitations, dimensionality reduction and UML methods show strong potential for performance analysts and coaches, enabling objective, data-driven identification of tactical tendencies to guide training, opponent analysis, and recruitment. Future research should expand to a wider range of sports, adopt standardised methodological frameworks, and integrate spatiotemporal tracking data for more detailed and interpretable analyses of playing style.