NBA Fantasy Score Projection

For this project we built a model that DFS players can utilize to pick better lineups for fantasy competitions. Trained on 5 seasons of NBA game data, the model outputs a prediction for a player's fantasy score in an upcoming game based on factors like the opposing team, the game being home or away, and his performances in both the short-term and relatively long-term past.

We compared the prediction accuracy on a test set after training a pair of random forest and a pair of XGBoost models in order to determine which gave the optimal performance. The winner was not actually technically a single model, but 7 independent ones that would each predict a single stat out of the 7 categories that mathematically factor into fantasy score. Then we simply applied the open-source DraftKings fantasy formula to those predictions to calculate the point value. When using these predictions in conjunction with salary cap optimization for a DraftKings contest, we achieved an 8.58% improvement in lineup performance over the DraftKings-provided point projections.