Key factors and tactical variations in Chinese national women's softball games: a machine learning-based identification.

Objective: This study employs machine learning to analyze data from Chinese women's softball games, identifying key factors determining game outcomes. It explores patterns in how different teams develop winning strategies. Method: This study analyzed data from 81 of 296 games conducted between 2023 and 2024, using game outcomes (win = 1, loss = 0) as the target variable and 98 features as inputs. Machine learning models, including Random Forest (RF), XGBoost, KNN, and SVM, were implemented in Python and trained on a 7:3 train-test split. Model performance was evaluated using AUC, F1-score, accuracy, precision, and recall to identify the best-performing model. SHAP and PDP were then employed to evaluate feature contributions to game outcome predictions. Results: The RF model achieved the highest accuracy on the test set with an AUC of 97.7% (95% CI: 0.938, 0.993). We identified the ten features that had the most significant impact on game results, including P-ER, OBP, RBI, and AVG. PDP analysis further revealed that an increase in P-ER and P-H significantly increased the probability of losing; improvements in OBP and AVG substantially increased the chances of winning. Different teams exhibited varying strategic emphases in their decisive factors: Team SC relied heavily on pitching performance, while SH, LN, and JS prioritized batting strategies. Conclusion: Feature importance analysis from the RF model indicates that P-ER and key batting metrics (e.g., OBP, AVG)are significantly associated with predicting game outcomes. These findings highlight their importance in predictive models, though further research is needed to confirm their practical impact.