Creating models to forecast the outcome of the men's 2024 collegiate basketball tournaments.
Datasets that was given:
data_path = '../data'
os.listdir(data_path)['2024_tourney_seeds.csv',
'Cities.csv',
'Conferences.csv',
'MConferenceTourneyGames.csv',
'MGameCities.csv',
'MMasseyOrdinals.csv',
'MNCAATourneyCompactResults.csv',
'MNCAATourneyDetailedResults.csv',
'MNCAATourneySeedRoundSlots.csv',
'MNCAATourneySeeds.csv',
'MNCAATourneySlots.csv',
'MRegularSeasonCompactResults.csv',
'MRegularSeasonDetailedResults.csv',
'MSeasons.csv',
'MSecondaryTourneyCompactResults.csv',
'MSecondaryTourneyTeams.csv',
'MTeamCoaches.csv',
'MTeamConferences.csv',
'MTeams.csv',
'MTeamSpellings.csv',
'sample_submission.csv',
'WGameCities.csv',
'WNCAATourneyCompactResults.csv',
'WNCAATourneyDetailedResults.csv',
'WNCAATourneySeeds.csv',
'WNCAATourneySlots.csv',
'WRegularSeasonCompactResults.csv',
'WRegularSeasonDetailedResults.csv',
'WSeasons.csv',
'WTeamConferences.csv',
'WTeams.csv',
'WTeamSpellings.csv']Datasets that I would be using:
-
MRegularSeasonCompactResults - Game results from the regular season
-
MRegularSeasonDetailedResults - Detailed Results
-
MNCAATourneyCompactResults - Game results from past tournaments
-
MNCAATourneySeeds - Seeding of past tournaments
-
2024_tourney_seeds - 2024 seeds
-
MTeams - Men's Team
season_results.columnsIndex(['Season', 'DayNum', 'WTeamID', 'WScore', 'LTeamID', 'LScore', 'WLoc',
'NumOT', 'WFGM', 'WFGA', 'WFGM3', 'WFGA3', 'WFTM', 'WFTA', 'WOR', 'WDR',
'WAst', 'WTO', 'WStl', 'WBlk', 'WPF', 'LFGM', 'LFGA', 'LFGM3', 'LFGA3',
'LFTM', 'LFTA', 'LOR', 'LDR', 'LAst', 'LTO', 'LStl', 'LBlk', 'LPF',
'League'],
dtype='object')-
Season - this is the year of the associated entry in MSeasons.csv or WSeasons.csv, namely the year in which the final tournament occurs.
-
DayNum - this integer always ranges from 0 to 132, and tells you what day the game was played on.
-
WTeamID - this identifies the id number of the team that won the game, as listed in the "MTeams.csv" or "WTeams.csv" file.
-
WScore - this identifies the number of points scored by the winning team.
-
LTeamID - this identifies the id number of the team that lost the game.
-
LScore - this identifies the number of points scored by the losing team. Thus you can be confident that WScore will be greater than LScore for all games listed.
-
WLoc - this identifies the "location" of the winning team.
-
NumOT - this indicates the number of overtime periods in the game, an integer 0 or higher.
-
WFGM - field goals made (by the winning team)
-
WFGA - field goals attempted (by the winning team)
-
WFGM3 - three pointers made (by the winning team)
-
WFGA3 - three pointers attempted (by the winning team)
-
WFTM - free throws made (by the winning team)
-
WFTA - free throws attempted (by the winning team)
-
WOR - offensive rebounds (pulled by the winning team)
-
WDR - defensive rebounds (pulled by the winning team)
-
WAst - assists (by the winning team)
-
WTO - turnovers committed (by the winning team)
-
WStl - steals (accomplished by the winning team)
-
WBlk - blocks (accomplished by the winning team)
-
WPF - personal fouls committed (by the winning team)
-
(and then the same set of stats from the perspective of the losing team: LFGM is the number of field goals made by the losing team, and so on up to LPF).
Creating a metrics of game results as "W" or "L"
team_results = pd.concat(
[season_results[["Season", "League","DayNum", "WTeamID", "WScore", "LScore"]].assign(GameResult="W")
.rename(columns={"WTeamID": "TeamID", "WScore": "TeamScore", "LScore": "OppScore"}),
season_results[["Season", "League","DayNum", "LTeamID", "WScore", "LScore"]].assign(GameResult="L")
.rename(columns={"LTeamID": "TeamID", "LScore": "TeamScore", "WScore": "OppScore"}),]
).reset_index(drop=True)
----------------------------------------------------------------------------------------------
team_results.sample(10, random_state=99) Season League DayNum TeamID TeamScore OppScore GameResult
2019 M 101 1410 59 44 W
2014 M 41 1461 61 64 L
2014 M 33 1420 68 70 L
2017 M 61 1223 51 61 L
2021 M 124 1455 80 63 W
2023 M 59 1158 89 61 W
2015 M 116 1156 53 56 L
2010 M 16 1428 68 67 W
2009 M 47 1253 61 75 L
2003 M 96 1225 44 52 LUsing merge() and groupby() of the TeamIDs and TeamNames
team_performance = team_results.groupby(['Season','TeamID','GameResult']).size().unstack(fill_value=0)
team_performance = team_performance[['W','L']]
team_performance = team_performance.merge(teams,on='TeamID')
team_performance = team_performance.groupby('TeamName').agg({'W':'sum','L':'sum'}).reset_index()TeamName W L
Abilene Chr 136 152
Air Force 282 352
Akron 440 235
Alabama 417 270
Alabama A&M 189 406
... ... ... ... ... ... ...
Wright St 382 285
Wyoming 310 342
Xavier 473 219
Yale 323 255
Youngstown St 231 397Top 10 of Most Wins and Losses:
For 2024 Season, the Top 10 of Most Wins:
2024 Number 1 Chalk Seeds for each region:
seeds_2024.query('Chalk_Seed == 1')TeamName Chalk_Seed Region
Connecticut 1 W
North Carolina 1 X
Purdue 1 Y
Houston 1 Z- Effective Field Goal Percentage - eFG%
- True Shooting Percentage - TS%
Calculating the total points given by Kaggle:
season_results[['Season','League','DayNum','WTeamID','LTeamID','ScoreDiff','TS%_Diff','eFG%_Diff']]Season League DayNum WTeamID LTeamID ScoreDiff TS%_Diff eFG%_Diff
2003 M 10 1104 1328 6 0.021201 0.057417
2003 M 10 1272 1393 7 0.081874 0.080886
2003 M 11 1266 1437 12 0.148889 0.160841
2003 M 11 1296 1457 6 0.092575 0.084586
2003 M 11 1400 1208 6 0.096529 0.105500
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
2024 M 114 1454 1237 5 0.055799 0.022289
2024 M 114 1455 1412 8 0.111259 0.126706
2024 M 114 1459 1359 22 0.158614 0.232512
2024 M 114 1462 1177 33 0.114577 0.143693
2024 M 114 1464 1178 14 0.122873 0.141778
112504 rows × 8 columnsGetting the total wins and losses as well as the margin wins and losses by using size() and mean()
Season TeamID Wins Losses
2003 1102 12.0 16.0
2003 1103 13.0 14.0
2003 1104 17.0 11.0
2003 1105 7.0 19.0
2003 1106 13.0 15.0
... ... ... ... ... ... ...
2015 1212 0.0 27.0
2015 1363 0.0 28.0
2021 1152 0.0 9.0
2022 1175 0.0 25.0
2022 1249 0.0 27.0
7617 rows × 4 columnsGetting the Win/Loss ratios and Overall Score
Season TeamID Wins AvgWinScoreDiff Losses AvgLossScoreDiff WinLossRatio AvgScoreDiff
2003 1102 12 15.583333 16.0 -11.250000 0.705882 0.250000
2003 1103 13 9.384615 14.0 -7.500000 0.866667 0.629630
2003 1104 17 13.176471 11.0 -9.454545 1.416667 4.285714
2003 1105 7 13.000000 19.0 -11.473684 0.350000 -4.884615
2003 1106 13 10.384615 15.0 -9.266667 0.812500 -0.142857
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
2024 1474 10 8.900000 18.0 -14.166667 0.526316 -5.928571
2024 1475 5 8.000000 21.0 -11.952381 0.227273 -8.115385
2024 1476 3 10.666667 25.0 -16.480000 0.115385 -13.571429
2024 1477 7 6.714286 18.0 -17.222222 0.368421 -10.520000
2024 1478 9 15.888889 16.0 -14.250000 0.529412 -3.400000
7609 rows × 8 columnsGetting the WinLoss Ratio
df['WinLossRatioDiff'] = df['WinLossRatio_W'] - df['WinLossRatio_L']
df['AvgScoreDiff'] = df['AvgScoreDiff_W'] - df['AvgScoreDiff_L']Inverting the data frame so the model gets examples to learn from
new_df = df.copy()
df_inverted = df.copy()
df_inverted['WinLossRatioDiff'] = -df['WinLossRatioDiff']
df_inverted['AvgScoreDiff'] = -df['AvgScoreDiff']
df_inverted['Outcome'] = 0
new_df['Outcome'] = 1
df = pd.concat([new_df, df_inverted], ignore_index=True)features = [
'DayNum', 'TS%_Diff', 'eFG%_Diff',
'WinLossRatio_W', 'WinLossRatio_L',
'AvgScoreDiff_W', 'AvgScoreDiff_L',
'WinLossRatioDiff', 'AvgScoreDiff'
]X = df[['WinLossRatioDiff', 'AvgScoreDiff']]
y = df['Outcome']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)- Random Forest Classifier
Random Forest Accuracy: 0.6674- Logistic Regression
Logistic Regression Accuracy: 0.7321- XGBoost Classifier
XGBoost Accuracy: 0.7356Did not complete bracket simulation in time but will use this as base for future use.