March Madness Data Crunch Overview

March Madness Data Crunch Overview

March Madness Data Crunch Overview Sponsored By: Prof. Petersen, Fordham University 1 Timeline 02/06/19 Historical Data Released 02/13/19 Registration Deadline for Teams 03/01/19 Initial Predictions CSV Submission due 03/18/19 2019 Current Season Data Released by 5PM 03/20/19 2019 Final Tournament Predictions CSV due by 5PM 03/27/19 2019 Final PowerPoint Report & Participation

Declaration Form due by 5PM 04/05/19 Final Poster Session & Awards Ceremony Prof. Petersen, Fordham University 2 Where to Create Teams Create a team of 4 and upload Excel file to blackboard by, 2/13/19, 11:59 pm Please email team name and members to: [email protected] Teams will then be added to the March Madness Blackboard class Materials will be uploaded there Prof. Petersen, Fordham University 3 Objective

Based on Kaggles Machine Learning Mania https://www.kaggle.com/c/march-machine-learning-mania-2017 Predict the probability that a team wins any given game in the March Madness Tournament Predict all possible 2278 matches Use data from 2002 until 2018 to train and test until data for 2019 is released Be creative! See if you can find signal in the noise Demonstrate your analytical skills Visualize your findings Prof. Petersen, Fordham University 4 Dataset Overview Glossary available on Blackboard Game Data: game_id, host name and latitude and longitude and score KenPom Data: four factor data, tempo, efficiency, etc.

Do not share outside of Fordham Coaching Data: Coach name, career wins, season wins, NCAA tournament appearances, Sweet 16 appearances, and Final 4 appearances Team Location Data: Latitude and longitude of team1 and team2 Team Data: Team Name Poll Data: AP Preaseason/Final Polls, Coaches Preseason/Final Polls RPI Data Prof. Petersen, Fordham University 5 Grading Criterion Judges will grade the submissions on the following factors Model Accuracy How well did the model perform? Creativity of Exploratory Analysis & Methodology Was the team able to find novel ways to improve accuracy and gain new insights into what makes teams succeed in March?

Communication & Visualization How well was the team able to effectively communicate their findings to the judges Extremely important to Deloitte!! Note: Model accuracy is not the most important. Very important to find creative ways to analyze the data and effectively communicate! Prof. Petersen, Fordham University 6 Format of Poster Board Overview & Introduction Hypothesis & Methodology Variable Selection, Analytics Explored, Data Mining Techniques Analytics & Results Results of Analytics, Results of Data Mining Techniques Conclusions & Suggestions for Improvement Performance of Model

Prof. Petersen, Fordham University 7 Tutorials What is Log Loss? (Blackboard) SPSS Logistic Regression Example (Blackboard) Python Logistic Regression Example (Blackboard) Method Software Link Decision Trees R http://statsguys.wordpress.com/2014/03/ Other examples (Right) Iterative Strength Rating

Python (NumPy) http://bit.ly/1vzPIJX Probability Distribution Excel http://bit.ly/1BvIadI Network Analysis STATA http://bit.ly/1GwgF2b Generalized Boosted Model (Decision Trees) R

https://github.com/chmullig/marchmania Ordinal Logistic Regression and ExpectationR http://bit.ly/1As9dEr Various http://bit.ly/1weYwQd SAS Proprietary (Interesting Video Discussion) SAS http://www.unf.edu/~jcoleman/dance.htm Ensemble (Support Vector, Nave Bayes, KNN, Decision Tree, Random Trees, Neural N/A Nets) http://bit.ly/1BvIskU

Random Forest Python (scikit-learn) http://bit.ly/1LINr2U Prof. Petersen, Fordham University 8 Prediction Tracking http://fordhamsportsanalytics.com/ Prof. Petersen, Fordham University 9

Recently Viewed Presentations