This repository contains a few models for predicting results of NCAA tournament games. It was done as the group project for EECS 649 (Introduction to Artificial Intelligence).
The polynomial_regression folder contains code for a 3rd-degree polynomial regression that predicts game scores, written in Python using scikit-learn. The logistic_regression folder contains logistic models written in Python - using scikit-learn - and R. Finally, the other_models folder contains code for a dense neural network classifier, a random forest classifier, and a k-nearest-neighbor classifier. All models were trained on NCAA regular season data and evalutated on NCAA tournament data.
The polynomial regression model achieved a mean absolute error of 7.83 points on the training set, and 8.20 points on the test set. That is, its predictions were on average around 4 points off from the team's true score. r-squared for this model was 0.265, and when applied to the classification task (predicting winners of games) it achieved 72.1% accuracy.
The logistic regresssion model achieved 73.4% accuracy on the training set (regular-season games) and 68.9% on NCAA tournament games. The R model, though, achieved 72.1% on the training set and 65.3% on the test set.
The neural network achieved 73.66% accuracy on the regular season data and 71.66% on tournament data.
Data was taken from the NCAA Tournament prediction competition on Kaggle and the following columns were extracted from the box score datasets:
TeamID: The integer ID of the team from the datasetSeason: The season (i.e. 2008)2ptpct: The percentage of two-point shots made by the team3ptpct: The percentage of three-point shots made by the teamFTpct: The team's free-throw percentageFPG: Average team fouls per gameBPG: Average blocks per gameSPG: Average steals per gameAPG: Average assists per gameORPG: Average offensive rebounds per gameDRPG: Average defensive rebounds per gamePPG: Average points per gameRPG: Average rebounds per game. That is, the sum of offensive and defensive reboundsOrdinal: The team's KenPom rating at the time of the gameFPG_diff: Average foul difference. That is, the average difference between the team's fouls and the opposing team's, computed over every game.BPG_diff: Average blocks per game differenceSPG_diff: Average steals per game differenceAPG_diff: Average assists per game differenceORPG_diff: Average offensive rebounds per game differenceDRPG_diff: Average defensive rebounds per game differencePPG_diff: Average points per game differenceRPG_diff: Average rebounds per game difference