Malaysia Super League Match Results Prediction with Football Rating System and Machine Learning Algorithms

This paper explores the application of football rating system and machine learning algorithms to predict match results in the Malaysia Super League (MSL). The dataset spans eight seasons of MSL, from 2015 to 2022, encompassing variables such as match round, match date, home and away team names, goals scored by each team, goal difference, and match outcomes (win, draw, lose). The dataset from seven seasons of MSL 2015 until MSL 2021 serves as a training set to predict match outcomes for the MSL 2022 season. The study implements a novel Football Rating System comprising two components: Elo rating and pi-rating. The Elo rating and pi-rating systems are utilized to evaluate team strengths and performance in terms of rating across the league. Machine learning algorithms, including Naïve Bayes (NB), Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF), are employed to predict match outcomes in terms of probability of win, draw, and lose. The performance of these algorithms is assessed using the Rank Probability Score (RPS) and Accuracy (ACC) metrics. A lower RPS indicates superior performance, while higher ACC values signify better predictive accuracy. The findings highlight that the combination of Elo rating with Logistic Regression (ELO + LR) achieves the most favorable outcome, attaining the lowest RPS score of 0.1979 in comparison to the other models. Additionally, Elo Rating with Random Forest (ELO + RF) exhibits the highest accuracy at 56.06%, albeit with the third highest RPS value. This underscores the intricate relationship between ACC and the RPS, suggesting the significance of a balanced evaluation of predictive models. In summary, this research highlights the possibilities of utilizing machine learning algorithms and an advanced football rating system for predicting match outcomes in the Malaysia Super League. The research contributes to the knowledge of predictive modeling in sports and offers valuable insights aimed at improving the accuracy of outcome predictions.