A Comparative Study of Ensemble Methods and Feature Selection Techniques for Predicting English Premier League Match Outcome

Accurately predicting the outcomes of football matches presents a complex challenge due to the dynamic and multifactorial nature of the sport. This study investigates the effectiveness of ensemble learning models with embedded feature selection – Random Forest, XGBoost, and LightGBM – compared to a traditional Support Vector Machine classifier combined with explicit feature selection techniques, including LASSO, PCA, RFE, and ANOVA. Using a constructed dataset based on pre-match statistics from the English Premier League, we evaluate each model’s performance in predicting home versus away wins. Results from experiments reveal that ensemble methods offer slightly superior performance. However, SVM models with feature engineering approaches such as LASSO and ANOVA also perform competitively. These findings suggest that both embedded and explicit feature selection strategies can be effective, and the choice of model may depend on practical considerations such as interpretability and computational cost.