Optimising Machine Learning Models to Predict the Outcome of Tennis Matches with a Focus on Objective Features
Yurui Zi
Issue:
Vol. 1 No. 1 (2023)
Date Published:
05-07-2023
Keywords:
Machine Learning Models, Neural Networks, Linear Regression, Profession Tennis, Betting Odds Prediction
ABSTRACT
The prediction of betting odds surrounding professional tennis matches is extremely complicated, making it perfect for analysis by means of training machine learning models. Billions of dollars are bet on tennis annually - however knowledge is limited with regards to how betting companies calculate odds of professional tennis matches. Previous research papers have utilised machine learning techniques which focus on player-centered features, such as serving and returning percentage, unforced errors and double faults to produce models that can predict the probability of a player winning a tennis match. In this paper, objective features of a tennis match which can be reliably measured before the commencement of the match, such as player ranking, court surface, location (indoor/outdoor), match format, and tournament round are analysed using two machine learning techniques - linear regression and neural networks. Using the ATP Tennis Dataset from Kaggle and through extensive analysis optimising hyperparameters, an accuracy of 67.41% was produced by a neural network of two hidden layers, each with 5 nodes. This accuracy was comparable to previous research. However, it can be concluded that the analysis of objective features is insufficient for an accurate prediction of gambling odds for a professional tennis match. Further research should be conducted with more features to inevitably synthesise both objective and subjective features to produce a more accurate prediction.