Traditional football analytics relies essentially on basic statistics such as goals and assists, which fail to capture the diverse contributions of midfielders on the field. This paper presents a new data-driven approach to identify and define midfielder roles using unsupervised machine learning techniques. We apply K-means clustering on performance metrics of players from Europe’s top five leagues from the 2020–21 FBRef dataset. Our methodology is to use 12 key performance indicators that describe many aspects of the midfielders’ contributions such as spanning ball progression, creativity, finishing, and defensive activities. We then identify four distinct midfielder archetypes by using Principal Component Analysis (PCA) and radar chart visualization: Deep-Lying Progressors, Defensive Midfielders, Wide Playmakers, and Advanced Playmakers. These data-driven roles can provide objective insights for player scouting, tactical analysis, and performance benchmarking, instead of the traditional and subjective role classifications.