Drivers of Winning in the NBA

Goal of Project

The goal of this project is to see what features help predict a team win using of NBA games played in 2014 through 2018. I will utilize the nba games.stats.csv found on Kaggle. A jupyter notebook will document all phases of the project including data acquisition, data preparation, exploration and modeling. A README.md will be made to describe and reproduce project.

Project Planning

Data Acquisition

• Download games.stats.csv on Kaggle.

Data Preparation

Some data preparation was performed prior to exploring the data.

• Dummy variables were created for categorical variables of home team and winorloss to have binary values (0 & 1).
• Some variable names were changed for easier interpretation (refer to the data dictionary).
• Split dataset into train, validate and test sets.
• Equations found in prepare.py.

Explore

• Created an explore.py file containing the recursive Feature Elimination function, Select K-Best function and scaled data frames to aid in modeling.

Model

• Classification models performed using model.py.
• Logistic Regression, Decision Tree, Random Forest and K-Nearest Neighbor algorithms performed.

Initial Thoughts

1. Team points and opponent points will tell me if a team wins or loses. I will not be using this in my analysis.
2. Field goal pct or field goals attempted will help in predicting a win.
3. 3 pt pct, assists, and total rebounds will help predict a win.
4. Home team will win more games than road team.

Hypothesis Testing

H₀: Does field goal percentage influence a win? Alpha < .05 so we reject the null hypothesis H₀.

H_a: Does field goals attempted influence a win? Alpha > .05 so we fail to reject the null hypothesis H₀.

H₀: Does 3 point shot percentage influence a win? Alpha < .05 so we reject the null hypothesis H₀.

H_a: Does assists influence a win? Alpha < .05 so we reject the null hypothesis H₀.

H_a: Does total rebounds influence a win? Alpha < .05 so we reject the null hypothesis H₀.

H_a: Does having home court influence a win? Alpha < .05 so we reject the null hypothesis H₀.

Data Dictionary

Dictionary

Column	Definition
team	Name of the home team
game	Game number in season
date	Date game played
opponent	Opponent team name
teampoints	Home team score
opponentpoints	Opponent score
fieldgoals	Number of field goals made by home team
fieldgoalsattempted	Number of field goals attempted by home team
fieldgoal_pct	Field goal percentage for home team
x3pointshots	3 point shots made by the home team
freethrows	Free throws made by the home team
freethrowsattempted	Free throws attempted by home team
freethrows_pct	Free throw percentage of home team
offrebounds	Number of offensive rebounds by the home team
totalrebounds	Total number of rebounds by the home team
assists	Number of assist by the home team
steals	Number of steals by the home team
blocks	Number of blocks by the home team
turnovers	Number of turnovers by the home team
totalfouls	Number of fouls by the home team
opp.fieldgoals	Opponents field goal percentage
opp.fieldgoalsattempted	Opponents field goals attempted
opp_fieldgoal_pct	Opponents field goal percentage
opp.3pointshots	Opponents 3 point shots made
opp.3pointshotsattempted	Opponents 3 point field goal percentage attempted
opp_3pointshots_pct	Opponents 3 point field goal percentage
opp.freethrows	Opponents free throws made
opp.freethrowsattempted	Opponents free throws attempted
opp_freethrows_pct	Opponents free throw percentage
opp.offrebounds	Opponents offensive rebounds
opp.totalrebounds	Opponents total rebounds
opp.assists	Opponents total number of assists
opp.steals	Opponents total number of steals
opp.blocks	Opponents total number of blocks
opp.turnovers	Opponents total number of turnovers
opp.totalfouls	Opponents total fouls
is_home	Away team = 0 Home team = 1

Target	Definition
win	Loss = 0 win = 1

Conclusions

• There are many features that can influence a win.
• Picked a variety of the top features found in RFE and SelectKBest.
• Feature in modeling include:
  • fieldgoals_pct, opp.fieldgoals_pct, opp.3pointshots, 3pointshots, opp.totalfouls, totalfouls, assist, and opp.assist
• Baseline accuracy of a win = 50%.
• Scaled trained data fit with Logistic Regression, Decision Tree, Random Forest, and K-NearestNeighbor.
• All models moved to validation phase.
• Logistic Regression accuracy performed the best with 87% on test.

  Model	Train	Validate	Test
  Logistic Regression	86%	86%	87%
  Decision Tree	82%	79%	NA
  Random Forest	85%	83%	NA
  KNN	88%	82%	NA

• Logistic Regression model accuracy was consistent through all phases of testing.
• Going foward, further testing can be done to optimize the best features to predict a win.
• Injuries play a big factor in the success of a team. I did not account for this in my models and information was not in data set. It would be interesting to see how models will perform with this data.

How to reproduce

• Download games.stats.csv on Kaggle.
• Download prepare.py, explore.py and model.py files to use in exploring and modeling data.
• Read through MVP_walkthru.ipynb to see analysis done.