A highly accurate Mamdani fuzzy inference system for tennis match predictions

Abstract This paper presents a Mamdani fuzzy inference system (FIS) designed for predicting tennis match outcomes with greater accuracy compared to existing models such as the Weighted Elo (WElo) ranking system. By integrating factors like historical performance, surface-specific proficiency, and recent form trends, the Mamdani FIS provides a nuanced approach to forecasting match results. Central to this method is the optimization of membership functions using a Bacterial Evolutionary Algorithm (BEA), which fine-tunes parameters to better model uncertainties inherent in sports analytics. This is the further development of Nawa and Furuhashi’s original approach of fuzzy system parameter discovery, which operates on the stricter conditions concerning the membership function shapes. The study demonstrates that the Mamdani FIS outperforms the traditional methods in both predictive accuracy and profitability of betting strategies. Through extensive validation, the model achieves higher accuracy and lower log loss metrics, indicating improved reliability in prediction outcomes. Additionally, the Mamdani FIS consistently yields higher returns on investment across various betting scenarios, showcasing its practical utility in sports betting applications. Overall, the proposed Mamdani FIS represents a robust tool for tennis match prediction, with potential extensions to other sports and predictive contexts. Future research may explore incorporating additional variables and applying this fuzzy inference approach to broader areas of sports analytics.