Implementasi Algoritma Gans Model Cgans pada Pemodelan Data Passing Networks: Pertandingan Liga Inggris 2024/2025

Abstract

Modern football tactical analysis requires sophisticated computational approaches to understand the complexity of game patterns. This research develops a Conditional Generative Adversarial Networks (CGAN) system to generate and analyze passing networks in Premier League season 2024/2025. The research objective is to implement an AI model capable of generating realistic passing networks based on specific tactical conditions, while providing an interactive visualization platform for in-depth analysis. The research methodology employs CGAN architecture with generator and discriminator optimized for football spatio-temporal data. Research data is obtained from Football-Data.org API and Fantasy Premier League, covering 380 matches with 579 players. The model is trained using PyTorch with tactical conditions including formations (4-3-3, 4-4-2, 4-2-3-1, 3-5-2, 5-3-2), match periods, and score situations. Implementation is complemented by a Streamlit-based interactive dashboard providing 6 comprehensive tactical analysis panels. Research results demonstrate that the CGAN model successfully achieves an average similarity score of 85% against actual data, with formation accuracy above 88% for all tested teams. Performance evaluation shows stable loss convergence after 350 epochs with discriminator accuracy reaching 92%. The tactical dashboard received user satisfaction ratings of 4.35/5 for visualization and 4.38/5 for analysis accuracy. The main contributions include developing CGAN architecture specific to football domain, integrating authentic Premier League data, and creating a user-friendly visualization platform for professional tactical analysis. The developed system provides practical solutions for coaches, analysts, and researchers to conduct tactical simulations and AI-based strategy exploration.