Machine Learning Approach for Malaysia Super League Football Match Outcomes Prediction Based on Elo Rating System

Predicting football match results or goals is unexceptionally buzzworthy. Football prediction can be classified into two clusters which are statistical and machine learning. Despite successfully introducing numerous statistical and machine learning techniques to predict football match outcomes, flaws still exist. This paper attempt to present football matches outcomes prediction models based on an Elo rating system and machine learning algorithms using limited data of football matches result for Malaysia Super League. The dataset used for the prediction is the MSL football data which consists of 7 seasons played between 2015 to 2021 that contain several basic features such as date, name of home team, name of away team, home team scored, away team scored and the matches results (Win, Draw, Loss). The football data were calculated to Elo rating that rate the strength of MSL football team before are divided into training and testing set. Machine learning (ML) algorithms such as Naïve Bayes (NB), Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) have been selected in this paper to predict football matches outcomes for MSL 2021. The accuracy and average of Rank Probability Score (RPS) are used as performance matric to evaluate the prediction models. Based on the comparative analysis conducted, all the models were able to predict the outcomes for more than 50% accuracy of the matches except RF which only obtained 49.24% accuracy. The NB is the best ML algorithm compared to SVM, LR and RF for predicting MSL football matches outcomes by achieved highest accuracy of 54.55% and lowest value of average RPS by 0.2025.