Comparing Machine Learning Methods for NBA Game Outcome Prediction

Basketball games generate a great amount of data across various dimensions, including player performance, team dynamics, and game outcomes. This makes basketball an ideal field for data analysis. In this study, we build upon previous researches to explore the prediction of NBA game outcomes based on key performance metrics such as points, rebounds and assists. We used several machine learning models, including logistic regression, decision trees, support vector machines, and random forests, to assess their prediction ability. We also used dimensionality reduction techniques like PCA and t-SNE to better understand the hidden structures of team performance. Our findings indicate that logistic regression with optimal regularization and random forest models achieved the highest accuracy and ROC AUC scores, demonstrating their potential for predicting win/loss outcomes. We also explored the impact of feature selection and hyperparameter, which provides valuable insights into our model's optimization. Although traditional machine learning techniques performed well, future research could add more performance factors, deal with imbalanced data, and explore deep learning approaches to further improve predictive accuracy.