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

Area of Science:

• Sports Analytics
• Machine Learning
• Basketball Performance Modeling

Background:

• Basketball analytics extensively studies game data, with machine learning advancing outcome prediction.
• Existing research primarily focuses on the NBA, leaving other major leagues underexplored.
• Predictive modeling for the Chinese Basketball Association (CBA) remains a significant research gap.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting Chinese Basketball Association game outcomes.
• To apply and adapt the Four Factors and Defense-Offense models, including detailed versions, to the CBA.
• To ensure practical applicability by using only pre-game data for model training and prediction.

Main Methods:

• Utilized CBA game data from the 2021-2024 seasons.
• Implemented classical Four Factors and Defense-Offense models, alongside detailed derivative versions.
• Trained and evaluated diverse machine learning algorithms (SVM, Naive Bayes, KNN, Logistic Regression, MLP, XGBoost) using pre-game data.
• Assessed model performance using Accuracy, F1 Score, Recall, Precision, and AUROC.

Main Results:

• The newly developed Four Factors detailed model, within a Logistic Regression framework, achieved the highest predictive performance.
• This model attained an accuracy of 85.49%, demonstrating substantial improvement over baseline approaches.
• Incorporating additional, detailed features significantly enhanced the predictive capabilities of all evaluated models.

Conclusions:

• The detailed Four Factors model offers a robust and accurate method for predicting CBA game outcomes.
• This research provides a valuable baseline for future sports analytics in the Chinese Men's Professional Basketball League.
• The findings underscore the importance of league-specific data and feature engineering for effective sports prediction models.