Adrian Pierce and Lotan Weininger's submission for the Google Cloud & NCAA® Machine Learning Competition 2019. This model placed 79th out of 868 teams (top 10%/Kaggle Bronze Medal).
Please see https://medium.com/@lotanweininger/march-madness-machine-learning-2dbacc948874 for a comprehensive write-up/article outlining the details of this submission and the competition experience.
Pre-requisites:
Download data files from https://www.kaggle.com/c/mens-machine-learning-competition-2019/data and place into "march-madness-2019/data"
- Calculate all required statistics:
1-stats-averages.rmd
1-stats-sd.rmd
1-stats-tempo.rmd
1-stats-pom.rmd
1-stats-rating.rmd
1-stats-possession.rmd
- Conduct exploratory data analysis:
2-eda.rmd
- Using the variables that best distinguish performant teams and non-performant teams, create training data:
3-create-training-data.rmd
- Test different combinations of the final variables with the logistic regression model (simulated competition results):
4-model-selection.rmd
- Using the model with the lowest error, find the optimal "clamp" value:
5-clamp-selection.rmd
- Merge training data into the correct submission format:
6-create-submission-data.RMD
- Create final model and write predictions:
7-create-final-model.Rmd
- Create list of unique teams in the 2019 tournament:
8-unique-teams.Rmd
- Manually label which teams are on "left side" of bracket vs "right side":
see "output/unique-teams-LR.csv" for the 2019 file
- Modify final submission files:
9-submission-LR.rmd
- Merge team names to final prediction file for readability:
10-predictions-named.Rmd
| Variable | Description | Team |
|---|---|---|
| X1 | Pomeroy Ranking | Team1 |
| X2 | Pomeroy Ranking | Team2 |
| X3 | Offensive Rating | Team1 |
| X4 | Offensive Rating | Team2 |
| X5 | Defensive Rating | Team1 |
| X6 | Defensive Rating | Team2 |
| X7 | Net Rating | Team1 |
| X8 | Net Rating | Team2 |
| X9 | Tempo | Team1 |
| X10 | Tempo | Team2 |
| X11 | Possession Time Per Game (sec.) | Team1 |
| X12 | Possession Time Per Game (sec.) | Team2 |
| X13 | Adjusted Pomeroy Ranking | Team1 |
| X14 | Adjusted Pomeroy Ranking | Team2 |
| Model | Variable Composition | Error |
|---|---|---|
| 1 | X1, X2, X3, X4, X5, X6, X7, X8, X9, X10 | 0.55346 |
| 2 | X3, X4, X5, X6, X7, X8, X9, X10 | 0.58481 |
| 3 | X1, X2, X3, X4, X5, X6, X7, X8 | 0.55322 |
| 4 | X1, X2, X7, X8 | 0.55342 |
| 5 | (X1 - X2), X7, X8 | 0.55345 |
| 6 | (X1 - X2), (X7 - X8) | 0.55291 |
| 7 | (X1 - X2), (X3 - X4), (X5 - X6), (X7 - X8) | 0.55257 |
| 8 | (X1 - X2)^3, (X3 - X4)^3, (X5 - X6)^3, (X7 - X8)^3 | 0.58617 |
| 9 | (X1, X2, X3, X4, X5, X6, X7, X8)^2 | 0.56862 |
| 10 | (X1 - X2), (X3 - X4), (X5 - X6) | 0.58856 |
| 11 | X3, X4, X5, X6 | 0.58472 |
| 12 | X1, X2, X3, X4, X5, X6, X7, X8, X11, X12 | 0.55551 |
| 13 | (X3 - X4), (X5 - X6), (X7 - X8) | 0.58462 |
| 14 | (X3 - X4), (X5 - X6), (X7 - X8), (X11 - X12) | 0.58793 |
| 15 | (X3 - X4), (X5 - X6), (X7 - X8), X13, X14 | 0.54982 |
| 16 | (X3 - X4), (X5 - X6), (X7 - X8), (X13 - X14) | 0.54966 |
| 17 | (X1 - X2), (X3 - X4), (X5 - X6), (X7 - X8), (X11 - X12) | 0.55587 |
| 18 | X1, X2, (X3 - X4), (X5 - X6), (X7 - X8) | 0.55329 |
Winning model: #16