Prediction methods comparison to determine LaLiga match outcomes

Predicting football match outcomes is a challenging task due to the stochastic nature of football and the limited information available prior to each match. This work analyzes and compares the predictive performance of different approaches for match outcome prediction using data from La Liga. A particular focus is placed on team rating systems, including the ELO-based method employed by FIFA, as well as alternative formulations that incorporate statistical performance indicators. In addition to rating-based approaches, statistical and machine learning models, including Poisson regression and gradient boosting methods, are evaluated. The results show that traditional rating systems provide a strong and interpretable baseline, outperforming some simple machine learning models. An enhanced rating system based on aggregated statistical performance achieves a notable improvement in predictive accuracy, while more advanced machine learning models offer moderate additional gains at the cost of reduced interpretability. Across all evaluated methods, predicting draws remains a significant challenge, highlighting the inherent uncertainty of football match outcomes. Overall, the findings suggest that no single approach is universally optimal and that model selection should depend on the intended application, balancing accuracy, interpretability, and robustness.