Key Factors Influencing NBA Game Outcomes: A Machine Learning Approach Using Game and Player Statistics

Predicting basketball game outcomes is complex as every game is influenced by many factors including individual players' performance and health conditions, team dynamics, team strategies, and game conditions. Previous studies have demonstrated growing interest in using machine learning techniques in sports analytics and have used only player performance statistics to predict the game results and produce the key predictors. This study incorporated additional factors such as game conditions and team momentum, in addition to those considered in previous studies. This study aimed to further develop a machine-learning approach to analyze key factors influencing NBA game outcomes in the 2024-25 season using game logs, player statistics, and aggregated historical performance data from the 2023-24 season. The study also optimized the XGBoost model using team-based train-test-validation split, five random seeds, feature selection, and hyperparameter tuning. The models were evaluated using various model performance metrics such as accuracy, F1 scores, specificity, sensitivity, and ROC-AUC. The findings indicated that win streaks, home-court advantage, field goal percentage, and past-season metrics such as steals and free throw percentage played significant roles in game outcomes. Additionally, SHAP values highlighted that win streaks and home-court advantages, rest days and travel schedules, and player trades and team performance had significant impact on the game result predictions. Then, individual player examples demonstrated how a player's performance and game condition influenced the game outcome through SHAP force plots. Future analysis should expand to variables such as ball possessions, player injuries, and team strategies and explore additional techniques such as controlled experiments or causal inference approaches to improve the model performance and provide specific and actionable recommendations.