Football Match Outcome Prediction Based on Team Ratings System

Abstract

Football outcome prediction is one of critical part in sports analytics that gives coaches, analysts, and fans a means to make strategic planning easier and improve engagement. In this study, a novel approach of combining Elo and Pi rating systems with machine learning models to predict the outcome of Malaysia Super League (MSL) matches is presented. Team performance can be evaluated by Elo and Pi ratings which are well known methods, integration which of the static and dynamic rating systems gives a more comprehensive assessment. Football historical match dataset from 2015 to 2022 was used to analyse machine learning algorithms such as Logistic Regression, Naive Bayes, K-Nearest Neighbors (KNN) and XGBoost. The performance of these algorithms was evaluated by using critical metrics such as accuracy, Ranked Probability Score (RPS), precision, recall, and F1-score. The Pi Rating combined with Naive Bayes achieves the highest accuracy and the best F1 Score across all metrics, making it the most superior combination for predicting match outcomes. Future research is recommended to apply k-fold cross validation, expand the dataset to include more diverse attributes and player specific statistics. By enhancing prediction accuracy and reliability, these enhancements can significantly improve sports analytics by better strategic decisions and team performance evaluations.