TITLE:
Comparative Evaluation of Machine Learning Models for NBA Game Outcome Prediction
AUTHORS:
Yuqi Wang
KEYWORDS:
NBA Game Prediction, Machine Learning, Feature Engineering, AutoGluon, Model Comparison
JOURNAL NAME:
Journal of Computer and Communications,
Vol.13 No.11,
November
12,
2025
ABSTRACT: This paper evaluates the performance of multiple machine learning models in predicting NBA game outcomes. Both regression and classification approaches were explored, with models including Logistic Regression, Lasso, Elastic Net, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Deep Neural Networks (DNN), and the AutoML framework AutoGluon. The results indicate that classification models outperform regression-based models, with SVM achieving the highest accuracy (0.7749), followed closely by AutoGluon (0.7738) and DNN (0.7726). Logistic Regression (0.7743) and Random Forest (0.7685) also demonstrated competitive performance, while KNN (0.7508) lagged behind. Feature engineering, such as standardization and polynomial expansion, significantly enhanced linear models, while regularization improved stability. Ensemble learning, particularly through AutoGluon, proved critical for capturing complementary model strengths, and deep learning benefited from dropout, batch normalization, and early stopping to mitigate overfitting. Overall, models that leverage non-linear structures and ensemble strategies showed the greatest effectiveness for sports outcome prediction.