Enhancing game outcome prediction in the Chinese basketball league through a machine learning framework based on performance data.

Basketball remains among the most globally popular sports, with its various competitions drawing substantial attention. The analysis and modeling of basketball game data have long been central topics in sports analytics. In recent years, integrating machine learning techniques has facilitated significant advancements in predicting basketball game outcomes. However, most existing studies predominantly focus on NBA data, with relatively limited exploration of other leagues. To address this research gap, this study utilizes game data from the Chinese Basketball Association spanning the 2021-2024 seasons to develop predictive models. This research is the first to apply the classical Four Factors model and DefenseOfense model, along with their derivative versions (Four Factors detailed model and DefenseOfense detailed model), to the Chinese Men's Professional Basketball League, providing a baseline for prediction. To ensure practical applicability of the models and enable their effective use in real-world scenarios, this study exclusively uses data available before the start of each game as feature variables for training. This approach ensures that the enhanced models can perform well in theoretical evaluations and provide reliable predictions when applied in practice. To evaluate model performance, a diverse set of machine learning algorithms, including support vector machines, Naive Bayes, k-nearest neighbors, logistic regression, multi-layer perceptron with contrastive loss, and XGBoost are employed, with metrics such as Accuracy, F1 Score, Recall, Precision, and AUROC used for comparison. The results reveal that the incorporation of additional features substantially enhances predictive performance. In particular, under the Logistic Regression framework, the newly developed model based on the Four Factors detailed achieves an accuracy of 85.49%, representing the highest predictive performance among all the evaluated approaches.