Predicting basketball game outcomes is a challenging task due to the sport’s dynamic nature, requiring rapid decision-making and strategic execution. Previous research has applied machine learning and reinforcement learning methods to predict game results, evaluate lineup performance, and determine optimal strategies, such as shot selection. While these studies have achieved significant predictive accuracy, they often lack interpretability. This work uses a Markov Decision Process (MDP) framework to model NBA games between fixed lineups, integrating historical NBA statistics and player-tracking data. By analyzing player interactions and decision outcomes, our model simulates half-court possessions, starting once the offensive team crosses half-court. The model estimates the probabilities of critical decisions, including passing, shooting, and rebounding. By simulating 100 possessions per team—typical for an NBA game—the model predicts game outcomes by comparing projected team scores. Model predictions are compared against actual NBA results and betting odds to evaluate predictive accuracy and profitability, applying betting strategies such as the Kelly Criterion. Our approach, leveraging data from the NBA SportVU dataset and NBA API, offers valuable insights into lineup performance.