Using Machine Learning to Predict NBA Games (Student Presentation, Group 22)

This is a STAT 451 class project presentation

by Almann Brague

This presentation is shared with the students' permission.

Abstract:

I use machine learning algorithms to predicted the out- come of NBA basketball games. As someone who has enjoyed the game of basketball all of my life, both as a participant and a fan. I hope through this project to find a machine learning algorithm to predict for wins and losses of NBA games. For this I have found statistics from NBA games from the years 2012 to 2018. I have assigned these statistics to each individual team to allow me to pre- dict which team would win against each other. I trained this on the years 2012-2017 and tested it on the most recent years of 2017-2018. The models that were used included, K-Nearest Neighbors (KNN), Decision Trees, Random Forests, and Gradient Boosting Classifiers. To find which model was best suited to create an accurate prediction I used a Nested Cross-Validation. This com- pared the four different algorithms together while also searching for their best hyper parameters. From this when apply the hyper parameters to each model that were best suited for prediction. I achieved prediction accuracy, on the training set, ranging from 57.04 percent to 65.76 per- cent. The best model from the the training set accuracy was the Gradient Boosting. Then when testing the Gradient Boosting method with specific hyper parameters on the test set it achieved a test accuracy of 65.89 percent.