XGBoost-Driven Evaluation of Clustering Methods for MOBA Player Segmentation

Authors : Sema Yildirim

Pages : 84-95

Doi:10.57020/ject.1770052

View : 49 | Download : 43

Publication Date : 2025-12-31

Article Type : Research Paper

Abstract :MOBA (Multiplayer Online Battle Arena) games are real-time, multiplayer digital games in which two teams compete strategically on a predefined map. Datasets from these games include diverse variables such as player performance, interactions, strategic behavior, and team dynamics. This study aims to cluster MOBA players based on behavioral patterns and assess the validity of these segments using classification techniques. During preprocessing, categorical variables were numerically encoded, and all features were scaled using the StandardScaler method. K-means and Hierarchical Clustering algorithms were applied to the dataset, and clusters were visualized in two dimensions via Principal Component Analysis (PCA). To evaluate cluster consistency, an XGBoost classification model was trained using stratified 5-fold cross-validation to predict cluster membership. The model achieved a mean accuracy of 0.91 ± 0.11 for the K-means clusters and 0.74 ± 0.07 for the Hierarchical clusters. These results indicate a strong internal structure in the K-means clusters. Additionally, the study explored relationships between player demographics (age, gender, weekly playtime) and psychological metrics (IGD-20, BIS-11, DERS) in relation to behavioral groupings. Findings highlight the value of combining unsupervised and supervised learning techniques to understand the complex profiles of MOBA players. Future work should focus on improving generalizability by incorporating larger, more diverse datasets and validating models with robust cross-validation protocols.
Keywords : Hierarchical clustering, player rank, K-means clustering, MOBA game, segmentation, XGBoost classification

ORIGINAL ARTICLE URL