The application of artificial intelligence techniques in predicting game outcomes of professional basketball league: A systematic review

Background: Predicting basketball game outcomes is a critical area in sports science and data analysis, providing concrete benefits for optimizing coaching strategies, improving team management, and informing betting decisions. Objective: This methodological review systematically evaluates the effectiveness of specific artificial intelligence technologies in predicting professional basketball game outcomes over the past five years from 2019 to 2024, providing detailed insights into current methodologies and identifying emerging trends and challenges in this domain. Methods: Following PRISMA-SCR guidelines, a comprehensive keyword search was conducted across four electronic bibliographic databases: PubMed, Web of Science, Scopus, and EBSCO. Studies were included if they utilized artificial intelligence techniques, focused on professional leagues, and aimed to predict game outcomes. Results: This review incorporated 34 studies that met the predefined eligibility criteria, examining various artificial intelligence techniques used to predict professional basketball game outcomes over the past five years. The findings reveal that artificial intelligence models, particularly the multilayer perceptron neural network, achieved a high prediction accuracy of 98.90%. The random forest model, based on four factors, reached an accuracy of 93.81%, while the voting regression ensemble model achieved 93.3%. The studies underscore the importance of effective data processing and feature selection in enhancing model performance. Additionally, dynamic prediction models that adapt to real-time changes in the game were shown to be particularly useful for tactical decisions and betting strategies. Conclusions: Artificial intelligence significantly improves the accuracy of predicting outcomes in professional basketball games. Future research should include diverse basketball leagues and employ more advanced validation techniques to enhance model robustness and applicability. Integrating real-time data and exploring transfer learning will likely improve prediction accuracy and decision-making support.