Athletic Prediction and Momentum Study Based on LightGBM Algorithm

In this paper, we investigate whether there are reliable metrics that can predict changes in state in competitive play. We constructed a score prediction model to assess performance. Firstly, principal component analysis (PCA) was used to distinguish between the score factor, process factor and skill factor; secondly, the LightGBM model was optimised using Bayesian optimisation to predict the score at a specific time, with a prediction accuracy of $82 \%$; then, using the predicted score results and the rest of the characteristics, a score factor evaluation system was established, and the indicators of the process and skill factors were weights, and established the evaluation system of process and technical factors respectively; the factor weights were obtained by AHP-TOPSIS method, and the performance scores were assessed by weighting the three factors. Finally, based on the score prediction model, we visualised and analysed the scores. These images demonstrate the correlation between the score and the performance score, and confirm that momentum significantly affects the change of the score.