Research on Dynamic Scoring Model of Soccer Game Based on XGBoost and LSTM

With the increasing complexity and real-time demand of soccer matches, traditional static scoring methods have significant limitations in capturing the dynamic changes and nonlinear feature relationships of the matches. To solve this problem, this paper proposes a dynamic scoring model for soccer matches based on the fusion of XGBoost and LSTM. By using XGBoost to model the static features of the game such as team's historical winning percentage, ranking, game environment, etc., and combining with LSTM to capture the time-series changes of dynamic features such as ball possession, number of shots, etc., during the game, the model realizes the synergistic modeling of static and dynamic information. The experimental results show that the fusion model significantly outperforms the traditional method and single model in terms of accuracy, F1 value and mean square error, and at the same time, it can reflect the impact of key events on the score during the game process in real time. This study not only provides an efficient technical tool for the dynamic scoring of soccer matches, but also provides new ideas for real-time analysis and dynamic prediction of other sports.