A new model for predicting the winner in tennis based on the eigenvector centrality.

Area of Science:

• Sports Analytics
• Network Science
• Statistical Modeling

Background:

• Statistical models for predicting tennis match winners are increasingly popular.
• Paired comparison models use latent abilities or ratings to estimate win probabilities.
• Existing models update player ratings only when they play matches.

Purpose of the Study:

• To extend paired comparison models using network indicators for enhanced prediction accuracy.
• To introduce a novel approach based on eigenvector centrality for player rating.
• To assess the performance of the new method against established models.

Main Methods:

• Proposed a new measure based on eigenvector centrality from network analysis.
• Utilized player ratings derived from centrality as a covariate in a logit model.
• Allowed player ratings to update dynamically with every new match, unlike traditional methods.

Main Results:

• The eigenvector centrality-based approach significantly outperformed competing models in prediction accuracy.
• The proposed method demonstrated consistent superiority across evaluations.
• The model also yielded positive results in betting scenarios.

Conclusions:

• Network indicators, specifically eigenvector centrality, provide a powerful tool for tennis match prediction.
• The dynamic updating of player ratings enhances predictive capabilities.
• This novel approach offers a significant advancement over existing statistical methods in sports analytics.