Wiki Lofandri, Geovanne Farell, Bayu Ramadhani Fajri, Fadli Ranuhardja, Agariadne Dwinggo Samala
In the competitive realm of Multiplayer Online Battle Arena (MOBA) games like Dota 2, support roles are often undervalued despite their strategic importance in determining match outcomes. Traditional metrics emphasize kills, gold accumulation, and damage output, which inadequately capture the tactical influence of support players. This study employs a deep learning approach—combining Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and XGBoost—to analyze a large-scale dataset from OpenDota comprising over 20 000 competitive matches. By leveraging match telemetry and spatio-temporal data, the proposed model quantifies support player impact using explainable AI (XAI) techniques such as SHAP. The results reveal distinct behavioral patterns and contributions of support roles in vision control, teamfight initiation, and strategic movement, offering a new lens for talent identification and coaching. This work contributes to the development of performance evaluation frameworks that holistically represent the complexity of team-based esports. © 2025; Los autores.
Universitas Negeri Padang, Faculty of Engineering, West Sumatra, Indonesia
