Football Match Prediction: A Monte Carlo Simulation Approach and Perspectives with Neural Networks

Abstract

Introduction: The unpredictability and low scoring in football make predictive modeling a high-variance challenge. The use of statistical databases, such as Football Manager, opens new avenues for sports analysis. Objective: This paper presents preliminary results of a heuristic Monte Carlo simulation applied to two knockout scenarios: the Campeonato Goiano (two-legged) and the 2026 World Cup playoffs (single-elimination), treated as independent case studies. Methodology: The Poisson distribution was used within 10,000 Monte Carlo iterations, with a dynamically modeled λ parameter based on each team’s relative strength from Football Manager 2026. Two pipelines were developed: an additive linear approach for the Goiano and a proportional ratio approach for the World Cup playoffs. Preliminary Results: The model achieved 83.3% accuracy (5/6) for the World Cup playoffs and 71.4% (5/7) for the Goiano, with Brier Scores of 0.2269 and 0.1229, respectively — both below the 0.25 randomguess threshold. These results support the progression towards a comparative study with Artificial Neural Networks and multiple linear regression in the full 2026 World Cup simulation.