Abstract
This study aims to analyze the key factors contributing to victories in world women's volleyball matches and predict match win rates using machine learning algorithms. Initially, Grey Relational Analysis (GRA) was employed to analyze the fundamental match data of the top six teams over three major world tournaments during the 2020 Olympic cycle (a total of 142 matches, 505 sets, and 27 metrics). The 27 metrics were used as subsequences, and the set win rate served as the parent sequence to identify metrics with a high contribution to match victories. Subsequently, the Gradient Boosting Decision Tree (GBDT) algorithm was utilized to construct a prediction model for match win rates, using the selected metrics as input features and set win rates as output features. The input metrics were ranked by their contribution to determine the most influential factors on match victories. The results indicate that spike scoring rate, blocking height, excellent defense rate, serve scoring rate, block scoring rate, proportion of serve scores, and proportion of block scores significantly impact match victories. Among these, spike scoring rate and blocking height are decisive, with feature importance values of 0.45 and 0.3, respectively. The constructed GBDT model demonstrated good predictive performance, capable of predicting match win rates. The model parameters are as follows: learning rate (learning-rate) of 0.1, number of trees (n-estimators) of 150, and maximum depth of the tree model (max-depth) of 2. The model's accuracy metrics on the test set are: MSE = 0.002, MAE = 0.0322, R(2) = 0.8497, and MAPE = 4.77%. The average relative error of the model validation is 5.30%, with R(2) = 0.743. This study not only identifies the key factors contributing to victories in world women's volleyball but also demonstrates the innovative application of combining Grey Relational Analysis with the Gradient Boosting Decision Tree algorithm in the volleyball domain. The findings provide data-driven insights to support coaches in training design and in-game decision-making, highlighting important practical value.