APPLICATION OF GENERALIZED LINEAR MODEL AND GRADIENT BOOSTING MACHINE TO PREDICT MATCH OUTCOME ON NCAA MEN’S BASKETBALL TOURNAMENT

• Main Author: Susanto, Marcello
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/49831
• Institution: Institut Teknologi Bandung

NCAA Men’s Basketball Tournament is a competition that bring the best men basketball university team in United States to compete each year. This tournament use a knockout format which mean that one lose can shatter the dream to win the championship, even the favorite one. For this reason each team must give their best effort on every match, which makes this competition even more competitive than regular season. The favorite team lose in the first match, the underdog can go deep to even win the tournament, all these scenario is very possible. This final paper will explore the ability of predictive model such as generalized linear model (GLM) and gradient boosting machine (GBM), and also find the most predictive variable using method such as GBM feature importance and GLM LASSO. The data use in this final paper consist of Las Vegas odds, Kenpom possession-based metrics, team statistics and team quality dataset which popularized by Kaggle user Darius Barusauskas (radar). Based on the logloss of the test dataset it can be concluded that GBM outperform GLM by small margin. The best model which have the lowest test logloss is GBM trained using regular season matchup on Las Vegas odds dataset that achieve test logloss of 0.52485