First Price in Fordham March Madness Data Crunch Competition, sponsored by Deloitte

Machine Learning, Python Sklearn, 2016 NCAA, Classification, Data Mining

Introduction

The National Collegiate Athletic Association (NCAA) Men's Basketball Tournament is informally referred to as "March Madness". With 68 college basketball teams competing in a single-elimination tournament, March Madness is played every spring in the US to determine the national championship of the major college basketball teams. The 68 teams are divided into four regions and respectively ranked 1 through 16 which determines the match-ups. Winning teams proceed through a single game elimination through 5 rounds. As a national tradition, “brackets” are filled in all over the country trying to predict the winners of each game until ultimately a champion of the tournament.

Objective

This project is to create an optimized model to predict 2016 NCAA Finals, based on historical regular season data from 2002 to 2015, through applying various machine learning techniques(such as Feature Engineering, PCA, Algorithm Optimization, Model Performance Validation and Model Fusion)

Result and Conclusion

Web API:

• http://shootingstarsnyc.azurewebsites.net/

Model Performance on training dataset:

• log loss: 0.52
• accuracy : 81%

Model performance on testing dataset: accuracy:

• Round of 32: 78%
• Sweet 16: 75%
• Elite Eight:75%
• Final Four: 50%
• Championship: 100%