Devin Powers

Devin Powers

Engineering Portfolio.

Machine Learning Project

Lets predict who will win the 2021 NCCA Basketball Tournament

I found a College Basketball Dataset (2013 - 2021 seasons on Kaggle) for their annual challenge.

Link to College Basketball Dataset

Here direct download to the College Basketball CSV file, using data from 2013-2019

CBB File Download

import matplotlib.pyplot as plt
import matplotlib
import pandas as pd
import numpy as np

Lets open the data set


column_names = ["Team", "Conference", "Games Played","Games Won",
          "Adjusted Offensive Efficiency", "Adjusted Defensive Efficiency",
          "Power Ranking", "Effective Field Goal %","Effective Field Goal % (D)",
          "Turnover %", "Turnover % (D)", "Offensive Rebounds", "Defensive Rebounds",
          "Free Throw Rate", "Free Throw Rate (D)", "2-PT%", "2-PT% (D)",
          "3-PT%", "3-PT (D)", "Adjusted Tempo", "Wins above Bubble", "Postseason","Seed in Tournament", "Year"]

data = pd.read_csv("cbb.csv", header = None, names = column_names, skiprows = 1)
data

data.head()

Output:

  Team Conference Games Played Games Won Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % … Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Postseason Seed in Tournament Year
0 North Carolina ACC 40 33 123.3 94.9 0.9531 52.6 48.1 15.4 … 30.4 53.9 44.6 32.7 36.2 71.7 8.6 2ND 1.0 2016
1 Wisconsin B10 40 36 129.1 93.6 0.9758 54.8 47.7 12.4 … 22.4 54.8 44.7 36.5 37.5 59.3 11.3 2ND 1.0 2015
2 Michigan B10 40 33 114.4 90.4 0.9375 53.9 47.7 14.0 … 30.0 54.7 46.8 35.2 33.2 65.9 6.9 2ND 3.0 2018
3 Texas Tech B12 38 31 115.2 85.2 0.9696 53.5 43.0 17.7 … 36.6 52.8 41.9 36.5 29.7 67.5 7.0 2ND 3.0 2019
4 Gonzaga WCC 39 37 117.8 86.3 0.9728 56.6 41.1 16.2 … 26.9 56.3 40.0 38.2 29.0 71.5 7.7 2ND 1.0 2017

Cleaning the Data

Lets drop all the teams that didn’t make the tournement.

In the Postseason Column teams that didnt make the tournemnt are read as a ??? marks (meaning missing) in the csv so we can use dropna() to remove those empty/missing rows.

data = data.dropna()

Theres also teams that lost their “play-in game” lets drop them as well

The tournemnt is 64 games and 68 with the play in games, In the CSV file teams with the Postseason = R68 didn’t win their play-in games. So we can single those rows out!! Lets update our data below:

data = data[data["Postseason"] != 'R68']
data

Output:

  Team Conference Games Played Games Won Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % … Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Postseason Seed in Tournament Year
0 North Carolina ACC 40 33 123.3 94.9 0.9531 52.6 48.1 15.4 … 30.4 53.9 44.6 32.7 36.2 71.7 8.6 2ND 1.0 2016
1 Wisconsin B10 40 36 129.1 93.6 0.9758 54.8 47.7 12.4 … 22.4 54.8 44.7 36.5 37.5 59.3 11.3 2ND 1.0 2015
2 Michigan B10 40 33 114.4 90.4 0.9375 53.9 47.7 14.0 … 30.0 54.7 46.8 35.2 33.2 65.9 6.9 2ND 3.0 2018
3 Texas Tech B12 38 31 115.2 85.2 0.9696 53.5 43.0 17.7 … 36.6 52.8 41.9 36.5 29.7 67.5 7.0 2ND 3.0 2019
4 Gonzaga WCC 39 37 117.8 86.3 0.9728 56.6 41.1 16.2 … 26.9 56.3 40.0 38.2 29.0 71.5 7.7 2ND 1.0 2017
… … … … … … … … … … … … … … … … … … … … … …
2450 Michigan St. B10 35 26 111.4 87.8 0.9392 50.6 44.5 20.8 … 32.4 50.4 44.3 34.1 30.1 64.4 6.7 S16 3.0 2013
2451 Arizona P12 35 27 114.4 92.2 0.9229 52.5 46.6 19.5 … 32.9 50.6 43.4 37.1 35.8 66.8 4.6 S16 6.0 2013
2452 Oregon P12 37 28 104.8 88.6 0.8728 49.3 46.4 21.4 … 33.3 49.1 44.9 33.3 33.4 69.2 2.9 S16 12.0 2013
2453 La Salle A10 34 24 112.0 96.2 0.8516 51.9 49.3 17.1 … 28.5 49.3 50.6 37.7 30.2 66.0 0.3 S16 13.0 2013
2454 Florida Gulf Coast ASun 35 24 103.4 96.3 0.6952 51.6 46.9 21.0 … 32.7 52.3 46.9 33.4 31.3 69.1 -4.0 S16 15.0 2013

448 rows × 24 columns

Now we can update the rows in Postseason to reflect the number of games each team won during the tournemnt in the year, the number of games won will be our Predictor (y)!

From the CSV file rows in the Column Postseason

  • Champions = 6 (Won the tournment)
  • 2ND = 5 (lost in the championship game)
  • F4 = 4 (Final Four)
  • E8 = 3 (Elite Eight)
  • S16 = 2 (Sweet Sixteen)
  • R32 = 1 (Round of 32)
  • R64 = 0 (Round of 64 the first round)

We can use .replace() and a dictionary to map/update the values to reflect the number of games won in each row!!

data = data.replace({
    
    'Champions': 6,
    '2ND': 5,
    'F4': 4,
    'E8': 3,
    'S16': 2,
    'R32': 1,
    'R64': 0
})

data

Output:

  Team Conference Games Played Games Wons Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % … Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Postseason Seed in Tournament Year
0 North Carolina ACC 40 33 123.3 94.9 0.9531 52.6 48.1 15.4 … 30.4 53.9 44.6 32.7 36.2 71.7 8.6 5 1.0 2016
1 Wisconsin B10 40 36 129.1 93.6 0.9758 54.8 47.7 12.4 … 22.4 54.8 44.7 36.5 37.5 59.3 11.3 5 1.0 2015
2 Michigan B10 40 33 114.4 90.4 0.9375 53.9 47.7 14.0 … 30.0 54.7 46.8 35.2 33.2 65.9 6.9 5 3.0 2018
3 Texas Tech B12 38 31 115.2 85.2 0.9696 53.5 43.0 17.7 … 36.6 52.8 41.9 36.5 29.7 67.5 7.0 5 3.0 2019
4 Gonzaga WCC 39 37 117.8 86.3 0.9728 56.6 41.1 16.2 … 26.9 56.3 40.0 38.2 29.0 71.5 7.7 5 1.0 2017
… … … … … … … … … … … … … … … … … … … … … …
2450 Michigan St. B10 35 26 111.4 87.8 0.9392 50.6 44.5 20.8 … 32.4 50.4 44.3 34.1 30.1 64.4 6.7 2 3.0 2013
2451 Arizona P12 35 27 114.4 92.2 0.9229 52.5 46.6 19.5 … 32.9 50.6 43.4 37.1 35.8 66.8 4.6 2 6.0 2013
2452 Oregon P12 37 28 104.8 88.6 0.8728 49.3 46.4 21.4 … 33.3 49.1 44.9 33.3 33.4 69.2 2.9 2 12.0 2013
2453 La Salle A10 34 24 112.0 96.2 0.8516 51.9 49.3 17.1 … 28.5 49.3 50.6 37.7 30.2 66.0 0.3 2 13.0 2013
2454 Florida Gulf Coast ASun 35 24 103.4 96.3 0.6952 51.6 46.9 21.0 … 32.7 52.3 46.9 33.4 31.3 69.1 -4.0 2 15.0 2013

448 rows × 24 columns

Sweet, now lets save that column as our Predictor (y)

y = data["Postseason"]
y

Output:

Index Postseason (Number of Wins)
0 5
1 5
2 5
3 5
4 5
..  
2450 2
2451 2
2452 2
2453 2
2454 2
Name: Postseason, Length: 448, dtype: int64  

Selecting the Feature Columns

Now the question is… What columns should we keep and what columns should we remove??

  • The number of games won in the regular reason will be different during our 2020-21 season since some teams werent able to play everyone, so I dont care about Games Played (record) and number of wins.

Lets Drop these:

We can only take Numeric Values as well

  • Team : Not Numeric Value (Can Drop)
  • Conference : Not Numeric Value (Can Drop)
  • Games Played : (Can Drop)
  • Games Won : (Can Drop)
  • Year : (Can drop)

Lets use these:

Note that the (D) stands for Defensively

  • Adjusted Offensive Efficiency
  • Adjusted Defensive Efficiency
  • Power Ranking : Odds of Beating a average D1 Team
  • Effective Field Goal %
  • Effective Field Goal % (D)
  • Turnover %
  • Turnover % (D)
  • Offensive Rebounds
  • Defensive Rebounds
  • Free Throw Rate
  • Free Throw Rate (D)
  • 2-PT%
  • 2-PT% (D)
  • 3-PT%
  • 3-PT (D)
  • Adjusted Tempo
  • Wins above Bubble
  • Postseason
  • Seed in Tournament

Lets drop those columns:

data = data.drop(columns = ["Postseason", "Team","Conference","Year","Games Played","Games Won"])
data.head()

Output:

  Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % Turnover % (D) Offensive Rebounds Defensive Rebounds Free Throw Rate Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Seed in Tournament
0 123.3 94.9 0.9531 52.6 48.1 15.4 18.2 40.7 30.0 32.3 30.4 53.9 44.6 32.7 36.2 71.7 8.6 1.0
1 129.1 93.6 0.9758 54.8 47.7 12.4 15.8 32.1 23.7 36.2 22.4 54.8 44.7 36.5 37.5 59.3 11.3 1.0
2 114.4 90.4 0.9375 53.9 47.7 14.0 19.5 25.5 24.9 30.7 30.0 54.7 46.8 35.2 33.2 65.9 6.9 3.0
3 115.2 85.2 0.9696 53.5 43.0 17.7 22.8 27.4 28.7 32.9 36.6 52.8 41.9 36.5 29.7 67.5 7.0 3.0
4 117.8 86.3 0.9728 56.6 41.1 16.2 17.1 30.0 26.2 39.0 26.9 56.3 40.0 38.2 29.0 71.5 7.7 1.0

Normalizing the Data

In general, learning algorithms benefit from standardization of the data set.

Standardization of datasets is a common requirement for many machine learning estimators implemented in scikit-learn; they might behave badly if the individual features do not more or less look like standard normally distributed data: Gaussian with zero mean and unit variance.

In practice we often ignore the shape of the distribution and just transform the data to center it by removing the mean value of each feature, then scale it by dividing non-constant features by their standard deviation.

data = ( data - data.mean())/data.std()

X = data

X

Output:

  Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % Turnover % (D) Offensive Rebounds Defensive Rebounds Free Throw Rate Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Seed in Tournament
0 1.852712 -0.267905 0.933216 0.166623 0.210833 -1.150197 -0.353803 2.097437 0.366500 -0.878417 -0.499655 0.910415 -0.629216 -1.279282 1.461359 1.292709 1.432846 -1.624281
1 2.784428 -0.522727 1.079968 0.961125 0.044102 -2.808673 -1.365528 0.067238 -1.815799 -0.154639 -1.868845 1.208926 -0.594961 0.216083 2.091614 -2.540566 2.000168 -1.624281
2 0.423009 -1.149982 0.832365 0.636102 0.044102 -1.924153 0.194215 -1.490822 -1.400123 -1.175351 -0.568114 1.175758 0.124406 -0.295489 0.006926 -0.500274 1.075643 -1.191268
3 0.551521 -2.169272 1.039886 0.491647 -1.914991 0.121301 1.585337 -1.042290 -0.083816 -0.767066 0.561467 0.545568 -1.554117 0.216083 -1.689913 -0.005658 1.096655 -1.191268
4 0.969187 -1.953653 1.060573 1.611173 -2.706964 -0.707937 -0.817510 -0.428509 -0.949807 0.364995 -1.098675 1.706445 -2.204972 0.885062 -2.029281 1.230882 1.243738 -1.624281
… … … … … … … … … … … … … … … … … … …
2450 -0.058914 -1.659627 0.843355 -0.555652 -1.289748 1.835059 0.067749 1.011516 -0.464852 -0.080406 -0.157357 -0.250462 -0.731983 -0.728358 -1.495989 -0.963977 1.033619 -1.191268
2451 0.423009 -0.797151 0.737979 0.130509 -0.414409 1.116386 0.320681 0.751840 -0.776609 0.068061 -0.071783 -0.184126 -1.040283 0.452193 1.267435 -0.222053 0.592368 -0.541749
2452 -1.119143 -1.502813 0.414091 -1.025131 -0.497775 2.166754 1.248095 0.940695 -0.603411 0.253645 -0.003324 -0.681645 -0.526450 -1.043171 0.103888 0.519871 0.235165 0.757289
2453 0.037471 -0.013082 0.277037 -0.086173 0.711027 -0.210394 0.953009 -0.664578 1.821366 -1.064000 -0.824837 -0.615309 1.426117 0.688303 -1.447508 -0.469361 -0.311146 0.973795
2454 -1.344040 0.006519 -0.734062 -0.194514 -0.289361 1.945624 1.290250 0.161666 1.336411 -0.340223 -0.106013 0.379728 0.158661 -1.003820 -0.914215 0.488958 -1.214659 1.406808

448 rows × 18 columns

Random Forest Classifier

  • One way to use predictive modeling is using Random Forest Classifier
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor

randTree = RandomForestRegressor(min_samples_split=20, random_state=5);
randTree.fit(X, y)

Output:

RandomForestRegressor(min_samples_split=20, random_state=5)

Now randTree is our trained model that we can input data (features X) and predict the number of wins that each team will win!

Testing our Model

Now that we have our Model we can import a file from 2018 and use that as a test to see how accurate our model is!!

Lets start again by importing data from 2019 College Basketball Season

2019 CBB File Download

Note: We removed Year from column_names since it’s not a column in the CSV file

column_names = ["Team", "Conference", "Games Played","Games Won",
          "Adjusted Offensive Efficiency", "Adjusted Defensive Efficiency",
          "Power Ranking", "Effective Field Goal %","Effective Field Goal % (D)",
          "Turnover %", "Turnover % (D)", "Offensive Rebounds", "Defensive Rebounds",
          "Free Throw Rate", "Free Throw Rate (D)", "2-PT%", "2-PT% (D)",
          "3-PT%", "3-PT (D)", "Adjusted Tempo", "Wins above Bubble","Seed in Tournament"]


test_data_2019 = pd.read_csv("cbb19.csv", header = None, names = column_names, skiprows = 1)
test_data_2019

Output:

  Team Conference Games Played Games Won Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % … Free Throw Rate Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Postseason Seed in Tournament
0 Gonzaga WCC 37 33 123.4 89.9 0.9744 59.0 44.2 14.9 … 35.3 25.9 61.4 43.4 36.3 30.4 72.0 7.0 E8 1.0
1 Virginia ACC 38 35 123.0 89.9 0.9736 55.2 44.7 14.7 … 29.1 26.3 52.5 45.7 39.5 28.9 60.7 11.1 Champions 1.0
2 Duke ACC 38 32 118.9 89.2 0.9646 53.6 45.0 17.5 … 33.2 24.0 58.0 45.0 30.8 29.9 73.6 11.2 E8 1.0
3 North Carolina ACC 36 29 120.1 91.4 0.9582 52.9 48.9 17.2 … 30.2 28.4 52.1 47.9 36.2 33.5 76.0 10.0 S16 1.0
4 Michigan B10 37 30 114.6 85.6 0.9665 51.6 44.1 13.9 … 27.5 24.1 51.8 44.3 34.2 29.1 65.9 9.2 S16 2.0
… … … … … … … … … … … … … … … … … … … … … …
348 Alcorn St. SWAC 27 10 89.0 112.6 0.0628 45.7 52.7 24.1 … 30.5 36.5 45.0 55.3 31.3 32.1 67.1 -16.7 NaN NaN
349 New Hampshire AE 27 5 83.7 106.1 0.0613 44.0 51.5 18.4 … 21.9 38.0 39.4 52.1 32.6 33.6 67.1 -20.2 NaN NaN
350 Chicago St. WAC 30 3 88.5 117.3 0.0380 44.2 57.8 22.5 … 33.1 33.9 43.5 57.9 30.7 38.5 71.9 -20.9 NaN NaN
351 Delaware St. MEAC 29 6 84.3 112.2 0.0358 40.0 52.4 19.0 … 25.5 39.2 37.7 52.6 29.0 34.7 71.6 -21.7 NaN NaN
352 Maryland Eastern Shore MEAC 30 7 85.7 114.4 0.0346 43.5 54.4 20.7 … 28.3 36.6 44.5 53.2 27.9 37.3 64.5 -19.9 NaN NaN

Then lets clean the data like we did above!!

# Lets Drop the teams that didn't make the tournment in 2018 and who ever didnt get into the round of 64

test_data_2019 = test_data_2019.dropna()

test_data_2019 = test_data_2019[test_data_2019["Postseason"] != 'R68']
 
## Lets replace the POSTSEASON with the number of wins

test_data_2019 = test_data_2019.replace({
    
    'Champions': 6,
    '2ND': 5,
    'F4': 4,
    'E8': 3,
    'S16': 2,
    'R32': 1,
    'R64': 0
})



# Lets grab the teams that made the Tournment
team_names = test_data_2019.get("Team")

actual_outcomes = test_data_2019.get("Postseason")

# Lets drop the Columns like we did above to the dataset
test_data_2019 = test_data_2019.drop(columns = ["Postseason", "Team","Conference","Games Played","Games Won"])

# Lets finally standardize the data! 

test_data_2019 = (test_data_2019 - test_data_2019.mean())/test_data_2019.std()

test_data_2019

Output:

  Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % Turnover % (D) Offensive Rebounds Defensive Rebounds Free Throw Rate Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Seed in Tournament
0 2.042572 -1.132484 1.073843 2.460166 -1.566239 -1.786817 0.010948 0.338659 -0.258338 0.390565 -1.264927 3.016431 -1.347685 0.262195 -1.161525 1.304614 0.951017 -1.614218
1 1.972384 -1.132484 1.068618 0.941607 -1.361502 -1.929228 -0.645904 0.058170 -0.736583 -1.103902 -1.164796 0.040748 -0.543228 1.656936 -1.875394 -2.729147 1.772165 -1.614218
2 1.252960 -1.260798 1.009833 0.302213 -1.238660 0.064530 0.186108 1.384121 0.663992 -0.115625 -1.740545 1.879653 -0.788062 -2.135016 -1.399481 1.875766 1.792193 -1.614218
3 1.463523 -0.857526 0.968031 0.022479 0.358290 -0.149087 -0.295583 1.307624 -1.624753 -0.838755 -0.639113 -0.092990 0.226254 0.218609 0.313805 2.732493 1.551857 -1.614218
4 0.498442 -1.920698 1.022243 -0.497028 -1.607186 -2.498873 -0.426954 -1.395277 -0.941545 -1.489571 -1.715513 -0.193294 -1.032897 -0.653104 -1.780211 -0.872903 1.391633 -1.398989
… … … … … … … … … … … … … … … … … … …
61 -1.905487 0.847215 -1.982937 -0.377142 0.726817 -0.077881 1.893923 -0.808798 0.322389 0.077209 1.839111 -0.962291 0.890806 0.872394 0.218622 -0.444539 -0.971671 1.398989
62 -1.010593 1.928718 -2.121406 0.781758 1.013449 -0.220292 -0.339373 -1.981755 1.278879 1.113695 -0.839373 0.408529 1.100664 0.872394 0.408988 0.126613 -1.452343 1.614218
63 -0.905312 2.368651 -2.436881 0.262251 2.282819 -1.217172 -2.003398 -2.415239 -0.565781 -0.356669 -1.114731 -0.092990 1.870145 0.436537 1.979500 -0.658721 -2.173351 1.614218
64 -1.150969 2.331990 -2.649158 0.062441 1.750502 -0.077881 -0.208003 -1.293281 1.073917 0.752130 -0.614080 0.274791 1.065688 -0.260833 1.789134 1.268917 -2.533854 1.614218
66 -1.273797 2.606948 -3.008395 0.541986 1.627660 1.631054 0.536429 -0.171322 1.927926 0.486983 -0.388787 -0.460771 1.415452 1.918449 1.170448 -0.123266 -2.313546 1.614218

64 rows × 18 columns

Now that we have everything formated, we can test our data on the train model that we build to test for accuracy!

outcomes = randTree.predict(test_data_2019)
outcomes

array([1.07743918, 1.13519988, 2.65392151, 1.41487909, 2.47988569,
       2.94260759, 1.10666421, 2.99517953, 1.47717599, 1.21751297,
       1.3863434 , 2.40559993, 0.87402628, 0.78681459, 1.36926636,
       2.35571116, 0.8465855 , 1.00924984, 2.56186303, 3.01548967,
       1.3497019 , 0.67274804, 2.4168949 , 0.97998758, 0.30464533,
       0.76005142, 0.3474276 , 0.70917375, 0.25876418, 2.25906946,
       1.4213046 , 1.08078719, 0.5007893 , 0.41390343, 2.62928611,
       0.53216457, 0.88700573, 0.03977468, 2.55307896, 0.77592342,
       0.86255786, 0.21795734, 0.38352635, 0.77422673, 1.01526413,
       0.77083375, 0.21385657, 2.55357066, 0.26152267, 0.37922084,
       0.08835621, 0.20948635, 2.53386749, 0.31673781, 0.52333739,
       0.04270588, 0.40829133, 2.66266546, 0.18301588, 0.15924488,
       0.10103648, 0.22302584, 0.04448503, 0.07210633, 0.27154693,
       0.10517421, 0.22233011, 0.05810671])

Lets look at the MSE and R Squared Values

R Squared: is the proportion of the variance in the dependent variable that is predictable from the independent variable(s).

  • Range from 0 to 1, closer to 1 is the best! ( 1 being Perfect)

MSE: Means Squared Error! (lower number the better)

The sklearn library has MSE and R Squared metrics build in, lets first import them!

from sklearn.metrics import mean_squared_error, r2_score

print('Mean squared error:', mean_squared_error(actual_outcomes, outcomes)) 
print('Coefficient of determination:', r2_score(actual_outcomes, outcomes))

Output:

Mean squared error: 0.3880797220485247
Coefficient of determination: 0.783995849774323

Are trained model looks pretty solid when testing the 2019 NCAA Data on it!

Running Model on 2021 Data

Lets now run our trained model on the 2021 Data to predict the number of wins for each team!!

2021 CBB File Download

We have to Change a few columns in the 2021 dataset!

column_names = ["Team", "Conference", "Games Played","Games Won",
          "Adjusted Offensive Efficiency", "Adjusted Defensive Efficiency",
          "Power Ranking", "Effective Field Goal %","Effective Field Goal % (D)",
          "Turnover %", "Turnover % (D)", "Offensive Rebounds", "Defensive Rebounds",
          "Free Throw Rate", "Free Throw Rate (D)", "2-PT%", "2-PT% (D)",
          "3-PT%", "3-PT (D)", "Adjusted Tempo", "Wins above Bubble", "Seed in Tournament"]

data_2021 = pd.read_csv("cbb21.csv", header = None, names = column_names, skiprows = 1)
data_2021

Output:

  Team Conference Games Played Games Won Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % … Defensive Rebounds Free Throw Rate Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Seed in Tournament
0 Michigan B10 24 20 118.1 91.1 0.9521 54.9 44.9 16.3 … 24.8 28.9 24.5 53.3 42.3 38.7 33.5 66.9 7.2 1.0
1 Baylor B12 24 22 123.2 94.5 0.9548 57.5 49.1 17.6 … 30.9 27.0 31.7 54.1 48.1 41.8 34.0 68.8 6.6 1.0
2 Illinois B10 29 23 117.7 90.4 0.9539 55.6 46.6 18.2 … 22.2 39.2 30.5 55.3 45.4 37.6 32.7 70.7 8.9 1.0
3 Gonzaga WCC 26 26 125.4 89.8 0.9791 61.0 47.5 16.1 … 23.4 36.7 25.9 64.0 46.8 36.5 32.5 74.6 8.5 1.0
4 Iowa B10 29 21 123.5 95.7 0.9491 54.6 48.3 13.3 … 28.6 32.0 22.6 52.4 45.8 38.6 34.8 70.0 5.6 2.0

Now we can Clean and Adjust the data!


# Drop teams that didn't make the tournament
data_2021 = data_2021.dropna()

#Grab Team names

team_names = data_2021.get("Team")

# Lets drop the Columns like we did above to the dataset

data_2021 = data_2021.drop(columns = ["Team","Conference","Games Played","Games Won"])

# Lets finally standardize the data! 

data_2021 = (data_2021 - data_2021.mean())/data_2021.std()

data_2021.head()

Output:

  Adjusted Offensive Efficiency Adjusted Defensive Efficiency Power Ranking Effective Field Goal % Effective Field Goal % (D) Turnover % Turnover % (D) Offensive Rebounds Defensive Rebounds Free Throw Rate Free Throw Rate (D) 2-PT% 2-PT% (D) 3-PT% 3-PT (D) Adjusted Tempo Wins above Bubble Seed in Tournament      
0 1.254529 -0.974462 0.932161 0.994790 -1.565422 -0.665798 -1.700853 -0.208926 -0.735535 -0.676980 -1.116763 0.508850 -2.126998 1.395668 0.642816 -0.302966 1.712498 -1.658834      
1 2.040990 -0.204484 0.948353 1.944973 0.797604 -0.067054 2.207902 1.853376 1.237944 -1.128067 0.244709 0.757337 0.360808 2.669843 0.907414 0.347254 1.544582 -1.658834      
2 1.192845 -1.132987 0.942956 1.250609 -0.608959 0.209289 -1.289405 0.707653 -1.576689 1.768387 0.017797 1.130067 -0.797309 0.943542 0.219460 0.997474 2.188261 -1.658834      
3 2.380248 -1.268865 1.094084 3.224066 -0.102596 -0.757912 0.438676 0.045679 -1.188464 1.174851 -0.852032 3.832363 -0.196804 0.491416 0.113621 2.332137 2.076317 -1.658834      
4 2.087253 0.067273 0.914169 0.885154 0.347504 -2.047515 -1.207115 0.122061 0.493845 0.059004 -1.476040 0.229302 -0.625736 1.354566 1.330770 0.757919 1.264722 -1.446003      

Now we can use our trained model to make Predictions (number of wins for each team)

outcomes = randTree.predict(data_2021)
outcomes

array([2.90137755, 3.25091381, 2.86165301, 4.76819429, 3.18066861,
       2.36148922, 3.09853133, 2.06465737, 1.92141687, 1.39338319,
       0.8818488 , 1.83575132, 2.52558497, 1.0863441 , 1.73735455,
       0.83906491, 1.74379691, 1.33183751, 1.12418422, 1.20195583,
       1.03637919, 1.4480753 , 1.11063843, 1.49674965, 0.64235945,
       1.30626018, 0.74411647, 0.47462401, 0.52989933, 0.79758753,
       1.2647636 , 1.4425606 , 0.63045054, 0.70913249, 2.06620984,
       0.86648306, 0.60994555, 0.61427192, 2.18634874, 0.52689784,
       0.33416316, 0.25236245, 1.0838805 , 0.79748643, 0.82304538,
       0.40971093, 1.07154771, 0.14501442, 0.28730544, 0.31200205,
       0.2381309 , 0.43729915, 0.18642513, 0.29956863, 0.46401838,
       0.10031835, 0.32652622, 0.08293184, 0.11272476, 0.1299452 ,
       0.15566156, 0.07748426, 0.08898297, 0.09433593, 0.16023765,
       0.05697938, 0.02390829, 0.09276995])

Convert Team names to a Frame

team_names = team_names.to_numpy()

Lets print out the Projected Wins for each team!

for i in range(len(team_names)):
    print(team_names[i], outcomes[i])

Projected Number of Wins for 2021 Tourny

Projected Number of Wins in Tourny for Teams

Michigan 2.901377551068698
Baylor 3.2509138093289214
Illinois 2.8616530060354
Gonzaga 4.768194292284091
Iowa 3.1806686060196636
Ohio St. 2.3614892221216515
Houston 3.098531328889728
Alabama 2.0646573670639774
West Virginia 1.9214168657035833
Texas 1.3933831859470671
Kansas 0.8818487965279312
Arkansas 1.83575131694493
Florida St. 2.5255849661183216
Virginia 1.08634409745
Purdue 1.737354550600605
Oklahoma St. 0.8390649078411792
Villanova 1.7437969052107938
Tennessee 1.3318375107722848
Creighton 1.1241842230904682
Colorado 1.2019558332965143
Texas Tech 1.036379192388832
BYU 1.4480752958728775
USC 1.1106384309237298
San Diego St. 1.4967496522059323
Florida 0.6423594519586673
Connecticut 1.3062601784053558
Clemson 0.7441164652698969
Oregon 0.47462400545930694
Oklahoma 0.5298993343723692
North Carolina 0.7975875291664838
LSU 1.2647635969699733
Loyola Chicago 1.4425605963594352
St. Bonaventure 0.6304505421114232
Missouri 0.709132494142787
Wisconsin 2.066209844623407
Georgia Tech 0.8664830586357855
Rutgers 0.6099455505034019
Virginia Tech 0.6142719174309312
Maryland 2.186348739514017
VCU 0.5268978448812078
Michigan St. 0.33416316257893075
Wichita St. 0.252362451365367
Syracuse 1.0838804985591546
UCLA 0.7974864348475628
Utah St. 0.8230453823210917
Drake 0.40971093402267866
Georgetown 1.0715477143484449
Oregon St. 0.14501442395206512
UC Santa Barbara 0.2873054411083263
Winthrop 0.31200205260351155
Ohio 0.23813090226222985
North Texas 0.4372991516463077
UNC Greensboro 0.18642513089080598
Liberty 0.29956862571123394
Colgate 0.4640183849244288
Eastern Washington 0.10031835221526803
Abilene Christian 0.3265262248702281
Morehead St. 0.0829318449854731
Iona 0.11272475691949374
Oral Roberts 0.12994520485084304
Grand Canyon 0.15566156131186967
Cleveland St. 0.07748425598062622
Drexel 0.08898296549941286
Mount St. Marys 0.09433592975651799
Hartford 0.16023764568764565
Norfolk St. 0.056979379782011355
Texas Southern 0.023908293460925042
Appalachian St. 0.09276995322684975

Important Features

What are the important Features?

Importances = randTree.feature_importances_
print('Features | Coefficients:')
print('-------------------------------')
for i in range(len(Importances)):
    print(data_2021.columns[i], ":", Importances[i])

Output:

Features | Coefficients:
-------------------------------
Adjusted Offensive Efficiency : 0.01929179106662344
Adjusted Defensive Efficiency : 0.012166986101116231
Power Ranking : 0.673670498282482
Effective Field Goal % : 0.016918569685206462
Effective Field Goal % (D) : 0.008659203235951584
Turnover % : 0.016029947205072663
Turnover % (D) : 0.011657031108119028
Offensive Rebounds : 0.04872405222299242
Defensive Rebounds : 0.03678456545563587
Free Throw Rate : 0.0287558387075352
Free Throw Rate (D) : 0.01523000282671945
2-PT% : 0.01392243965528682
2-PT% (D) : 0.013467855281397186
3-PT% : 0.014077173558140924
3-PT (D) : 0.016480450413741625
Adjusted Tempo : 0.015892583071390052
Wins above Bubble : 0.018270456403570653
Seed in Tournament : 0.020000555719018376

We can see above that some important Features (stats) are Power Ranking, Rebounding (both offensive and defensive, Free Throws

From here I am just going to look at the matchups on the NCAA Bracket and pick the winners of each game based on the the Projected wins, so whoever has the Higher Projected wins will advance in each matchup.

Bracket Filled Out

insert image

Some notably upsets include:

West

  • Missouri over Oklahoma in the 1st round
  • USC over Kansas in the 2nd round
  • VCU over Oregon in the 1st round
  • Creighton over Virgina (returning champions) in the 2nd round

East

  • Maryland over Uconn in the first round
  • Maryland over Alabama in the 2nd round
  • BYU over Texas in the 2nd round
  • Maryland making it to the Elite 8

South

  • Wisconsin over UNC (not really suprising)
  • Villanova over Purdue in the 2nd round

Midwest

  • Tennessee over OSU in the 2nd round
  • Houston over Illinois in the Elite 8 to make it to the Final 4

Final Four Projections

  • Gonzaga over Michigan

  • Baylor over Houston

Championship

  • Gonzaga over Baylor for the Title!

And the Winner of the 2021 NCAA Division 1 Basketball is…

Gonzaga!!!

insert image

THE GOAT ADAM MORRISON