The video game industry, particularly electronic sports, has experienced significant growth, with League of Legends emerging as a leading Multiplayer Online Battle Arena game. Success in professional matches heavily depends on strategic character selection during the draft phase. This paper uses data mining to identify and group similar characters’ play styles in professional League of Legends games. The proposed method refines traditional game roles into specific categories based on actual gameplay. The solution combines feature engineering, supervised classification, latent feature extraction with Shapley Additive Explanations, dimensionality reduction, and clustering. Results show that this approach performs better than traditional clustering methods, as indicated by higher silhouette indices and meaningful cluster distinctions. The solution aims to provide insights into character usage and potential replacements during drafts.