Application of K-Means Clustering Method for the Premier League Team Performance Segmentation

Final league table positions often fail to fully reflect a club's actual performance and playing style over the long term in competitions such as the Premier League. This study aims to group Premier League teams into homogeneous performance segments and present a comprehensive competitive landscape map. The problem is solved using a quantitative unsupervised learning approach by applying the K-Means Clustering algorithm to historical match statistics data spanning 12 seasons (2014/2015 to 2025/2026), comprising 4,560 official match records. Based on the Elbow method and Silhouette Score evaluations, the optimal number of clusters was determined to be . The input feature vector passed to the model strictly consisted of normalized Points and Goal Difference (GD) per season. The results successfully categorized Premier League clubs into three distinct strength strata: Elite (top-tier dominant teams), Competitive (upper-midtable contenders), and Vulnerable (relegation-prone teams), yielding a Silhouette Score of 0.450 and a Davies-Bouldin Index (DBI) of 0.741. Furthermore, baseline comparisons with Hierarchical Agglomerative Clustering (HAC) and Gaussian Mixture Models (GMM) confirmed 100% assignment convergence for the Elite stratum and demonstrated that K-Means achieved the most balanced cluster separation, providing deep strategic insights for football analysts to understand long-term club performance fluctuations.