Predicting Professional League of Legends Match Outcomes Using Supervised Machine Learning

Shreshta Adyaksa Hardono, Drs. Edi Winarko, M.Sc.,Ph.D

2025 | Skripsi | ILMU KOMPUTER

Prediksi hasil pertandingan League of Legends tingkat profesional memiliki peran penting dalam analisis esport, pencarian talenta, serta pengembangan strategi kompetitif. Meskipun sejumlah penelitian terdahulu telah memanfaatkan algoritma pembelajaran mesin yang kompleks, masih sedikit studi yang melakukan perbandingan sistematis antar berbagai model dalam konteks ekosistem profesional. Penelitian ini bertujuan untuk mengisi kesenjangan tersebut dengan mengevaluasi tiga arsitektur model—Logistic Regression, Random Forest, dan XGBoost—untuk mengidentifikasi pendekatan yang paling efektif dan sesuai tingkat kompleksitasnya dalam memprediksi hasil pertandingan. Menggunakan data pertandingan dari liga profesional utama, kinerja model dinilai berdasarkan akurasi validasi dan log loss.

Predicting the outcomes of professional League of Legends matches has significant implications for esports analytics, talent scouting, and competitive strategy development. While prior studies often rely on a single complex algorithm, limited research has directly compared multiple supervised machine learning models within the unique context of professional play. This study addresses that gap by systematically evaluating three model architectures—Logistic Regression, Random Forest, and XGBoost—to identify the most effective and appropriately complex approach for outcome prediction. Using match data from major professional leagues, each model’s predictive performance is assessed based on validation accuracy and log loss. In addition, the study investigates the transferability of predictive signals by applying the best-performing model to a dataset of Korean Challenger-tier ranked matches. This dual analysis enables a comparison of both predictive performance and feature importance between professional and high-elo ranked environments. The results indicate that while ensemble methods provide marginal improvements in log loss, simpler linear models achieve competitive accuracy with greater interpretability. These findings establish a clear methodological benchmark for esports outcome prediction and offer insights into the underlying statistical patterns that define victory across competitive tiers.

Kata Kunci : esports, league of legends, machine learning, predictive analysis, logistic regression