Data and scripts for competing in Google Cloud & NCAA® ML Competition 2019-Men's collaboratively with Evan Olawsky. Implemented predictive models such as penalized logistic and multiple linear regression, principle components regression, random forests, linear discriminant analysis, gradient boosting, support vector machines, and neural networks using R. Placed 483rd out of 866 submissions. A presentation summarizing our results can be found here.
Below is a bar plot displaying the performance of our machine learning models for predicting outcomes of games of the 2019 Men's NCAA Basketball Tournament using either a binary response (Win/Loss) or a continuous response of margin of victory (MOV) for model training and selection. Our model submitted for the competition was a linear discriminant analysis (LDA) model using a binary response since this performed the best on our test set in terms of classification accuracy. However, the single hidden layer neural network (NNet) using a Win/Loss response performed the best on the tournament data in terms of log-loss and classification accuracy.