Search papers, labs, and topics across Lattice.
This paper introduces an isotonic Bradley-Terry model that addresses limitations in traditional paired comparison methods by learning both rate parameters and an inverse link function using an alternating (sub-)gradient approach. The model ensures monotonic improvement in training error and can handle ties effectively when data is sparse, enhancing win probability predictions and player rankings. Empirical evaluations on datasets from football, baseball, and tennis demonstrate significant improvements over existing models in both prediction accuracy and ranking performance.
Traditional paired comparison models struggle with ties and misspecification, but this isotonic approach guarantees improved predictions and rankings even with limited data.
In this paper, we study prediction problems for paired comparison data, for example, predicting the win probability between two unmatched players and ranking all the players according to the order of their strengths by using win probability data between two matched players. Paired comparison data are typically analyzed using Bradley-Terry and Thurstone-Mosteller models. These models predict the win probability by transforming the difference between learned rate parameters, which represent players'\;strengths, with a pre-specified inverse link function, and employ the order of learned rate parameters for player ranking. However, these models may suffer from model misspecification owing to the selection of a fixed inverse link function. Therefore, in this study, we propose to learn the rate parameters by a (sub-)gradient method and the inverse link function by an isotonic regression technique alternately. The proposed model guarantees monotonic improvement in training error, and is likely to yield an exact tie when the available data is insufficient to establish a strict ranking. We also verified that the proposed model could improve the win probability prediction and ranking performance through numerical experiments with synthetic data and real-world data of football Premier League, baseball MLB, and tennis ATP tour.
