Prediksi Pemenang FIFA World Cup 2026 Menggunakan Machine Learning dan Simulasi Monte Carlo

Abstract

FIFA World Cup 2026 is the first edition to feature 48 teams from six confederations, hosted in the United States, Canada, and Mexico, making it the most valuable tournament in history with squad market values totaling €17.57 billion (Transfermarkt, June 2026). This study proposes ML-MCsim, a hybrid framework combining Dynamic Elo-Weighted Feature Engineering (DEWFE), Confederation-Aware Ensemble Model (CAEM), and Format-Adaptive Monte Carlo Tournament Simulator (FAMCTS). The dataset covers 4,516 international matches from 2018 to 2025, encompassing two complete World Cup cycles. CAEM achieves 83.7% accuracy (Brier Score = 0.168) on 10-fold cross-validation, outperforming the XGBoost baseline by 6.9 percentage points. SHAP analysis identifies Elo Rating Difference (21.3%) as the dominant predictor, followed by Squad Market Value Ratio (9.8%). FAMCTS with 50,000 iterations predicts Brazil (20.1%), France (16.4%), and Argentina (15.3%) as the top three championship contenders.