The goal of this project is to build a linear regression model that can predict the number of wins for NBA teams based on the point differential for each team. The model will be trained on data from the 1980-2011 NBA seasons and tested on data from the 2022-2023 NBA season.
The dataset used for this project is available in two files:
NBA_train.csv: Training data from the 1980-2011 NBA seasons.NBA_test.csv: Test data from the 2022-2023 NBA season.
To run the code, you will need to install the following packages:
numpypandasmatplotlibseabornscikit-learn
You can install these dependencies using the requirements.txt file:
pip install -r requirements.txtThe code begins by importing the required libraries.
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn import metricsWe load the training and test data from the CSV files, followed by providing an overview of the dataset, including the first 5 rows, data information, and summary statistics.
# Load the training and test data
try:
data_train = pd.read_csv('NBA_train.csv')
data_test = pd.read_csv('NBA_test.csv')
except FileNotFoundError:
print("Error: Data files not found.")
exit(1)
# Display basic information about the dataset
print("First 5 rows of the training dataset:")
print(data_train.head())
print("\nInformation about the training dataset:")
print(data_train.info())
print("\nSummary statistics of the training dataset:")
print(data_train.describe())The output of the above code is shown below:
First 5 rows of the training dataset:
SeasonEnd Team Playoffs W ... AST STL BLK TOV
0 1980 Atlanta Hawks 1 50 ... 1913 782 539 1495
1 1980 Boston Celtics 1 61 ... 2198 809 308 1539
2 1980 Chicago Bulls 0 30 ... 2152 704 392 1684
3 1980 Cleveland Cavaliers 0 37 ... 2108 764 342 1370
4 1980 Denver Nuggets 0 30 ... 2079 746 404 1533
[5 rows x 20 columns]
Information about the training dataset:
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 835 entries, 0 to 834
Data columns (total 20 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 SeasonEnd 835 non-null int64
1 Team 835 non-null object
2 Playoffs 835 non-null int64
3 W 835 non-null int64
4 PTS 835 non-null int64
5 oppPTS 835 non-null int64
6 FG 835 non-null int64
7 FGA 835 non-null int64
8 2P 835 non-null int64
9 2PA 835 non-null int64
10 3P 835 non-null int64
11 3PA 835 non-null int64
12 FT 835 non-null int64
13 FTA 835 non-null int64
14 ORB 835 non-null int64
15 DRB 835 non-null int64
16 AST 835 non-null int64
17 STL 835 non-null int64
18 BLK 835 non-null int64
19 TOV 835 non-null int64
dtypes: int64(19), object(1)
memory usage: 130.6+ KB
None
Summary statistics of the training dataset:
SeasonEnd Playoffs ... BLK TOV
count 835.000000 835.000000 ... 835.000000 835.000000
mean 1996.319760 0.574850 ... 419.805988 1302.837126
std 9.243808 0.494662 ... 82.274913 153.973470
min 1980.000000 0.000000 ... 204.000000 931.000000
25% 1989.000000 0.000000 ... 359.000000 1192.000000
50% 1996.000000 1.000000 ... 410.000000 1289.000000
75% 2005.000000 1.000000 ... 469.500000 1395.500000
max 2011.000000 1.000000 ... 716.000000 1873.000000
[8 rows x 19 columns]
We create a new column PointDiff by subtracting the opponent's points (oppPTS) from the team's points (PTS).
# Create a new column 'PointDiff' by subtracting 'oppPTS' from 'PTS'
data_train['PointDiff'] = data_train['PTS'] - data_train['oppPTS']We visualize the relationship between the point differential (PointDiff) and the number of wins (W) for each team in
the training set.
# Visualize the relationship between 'PointDiff' and 'W' on the training set
plt.figure(figsize=(15, 10))
sns.scatterplot(x='PointDiff', y='W', data=data_train)
plt.title('Point difference vs. Win totals (Training Set)')
plt.xlabel('PointDiff')
plt.ylabel('W')
plt.savefig('PointDiff_vs_W_Training.png')
plt.show()
We prepare the data for regression by selecting 'PointDiff' as the feature (X) and 'W' as the target (y). We also split the data into a training and a test set.
# Prepare the data for regression
X = data_train[['PointDiff']]
y = data_train['W']
# Split the data into a training and a test set
X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.7, random_state=42)We create and train a linear regression model, display its coefficients and intercept, and calculate the R-squared ( training set) to assess its performance.
# Create and train the linear regression model
model = LinearRegression()
model.fit(X_train, y_train)
# Display the model's coefficients, intercept, and equation
print("\nModel Coefficients:", model.coef_)
print("Model Intercept:", model.intercept_)
print("Regression Equation: W = {:.5f} * PointDiff + {:.5f}".format(model.coef_[0], model.intercept_))
# Evaluate the model on the training set
training_r2 = model.score(X_train, y_train)
print("\nR-squared (training set):", training_r2)
# Predict 'W' values on the test set
y_pred = model.predict(X_test) The output of the above code is shown below:
Model Coefficients: [0.03219151]
Model Intercept: 41.0766448346633
Regression Equation: W = 0.03219 * PointDiff + 41.07664
R-squared (training set): 0.9362053691644487
We visualize the model's performance on both the training and test sets using scatter plots and regression lines.
# Visualize the results on the training set
plt.figure(figsize=(15, 10))
sns.scatterplot(x='PointDiff', y='W', data=data_train)
plt.plot(X_train, model.predict(X_train), color='red')
plt.title('Point difference vs. Win totals (Training Set)')
plt.xlabel('PointDiff')
plt.ylabel('W')
plt.savefig('PointDiff_vs_W_Training.png')
plt.show()
# Visualize the results on the test set
plt.figure(figsize=(15, 10))
sns.scatterplot(x='PointDiff', y='W', data=data_train)
plt.plot(X_test, y_pred, color='red')
plt.title('Point difference vs. Win totals (Test Set)')
plt.xlabel('PointDiff')
plt.ylabel('W')
plt.savefig('PointDiff_vs_W_Test.png')
plt.show()
We evaluate the model on the test set by calculating the R-squared, MAE, MSE, RMSE, and cross-validation scores.
# Evaluate the model on the test set (1st evaluation)
test_r2 = model.score(X_test, y_test)
mae = metrics.mean_absolute_error(y_test, y_pred)
mse = metrics.mean_squared_error(y_test, y_pred)
rmse = np.sqrt(mse)
cv_scores = cross_val_score(model, X, y, cv=5)
print("\nR-squared (test set):", test_r2)
print("Mean Absolute Error (MAE):", mae)
print("Mean Squared Error (MSE):", mse)
print("Root Mean Squared Error (RMSE):", rmse)
print("\nCross-validation scores:", cv_scores)
print("Average cross-validation score:", cv_scores.mean())The output of the above code is shown below:
R-squared (test set): 0.9542929196391755
Mean Absolute Error (MAE): 2.2678549517024127
Mean Squared Error (MSE): 7.914354249601504
Root Mean Squared Error (RMSE): 2.813246212047837
Cross-validation scores: [0.93355201 0.93572823 0.94524537 0.9477965 0.94470755]
Average cross-validation score: 0.9414059318843483
We predict the number of wins for each team in the 2022-23 NBA season using the model we trained earlier.
# Add 'PointDiff' column to the test data, predict 'W' values, and round them
X = data_test[['PointDiff']]
y_pred = model.predict(X)
data_test['W_pred'] = y_pred.round(0).astype(int)We visualize the relationship between the point differential (PointDiff) and the number of wins (W) for each team in
the 2022-23 NBA season, as well as the actual and predicted win totals for each team.
# Display the predicted win totals for the 2022-23 season
plt.figure(figsize=(15, 10))
sns.scatterplot(x='PointDiff', y='W', data=data_test)
plt.plot(X, y_pred, color='red', linewidth=3)
plt.title('Point difference vs. Win totals (2022-23 Season)')
plt.xlabel('PointDiff')
plt.ylabel('W')
plt.savefig('PointDiff_vs_W_2022-23.png')
plt.show()
# Display the actual and predicted win totals for the 2022-23 season
df = data_test[['Team', 'W', 'W_pred']]
plt.figure(figsize=(15, 10))
plt.bar(df['Team'], df['W'], label='Actual W', alpha=0.7)
plt.bar(df['Team'], df['W_pred'], label='Predicted W', alpha=0.7)
plt.xlabel('Team')
plt.ylabel('Win Totals')
plt.title('Actual vs. Predicted Win Totals for the 2022-23 Season')
plt.xticks(rotation=90) # Rotate x-axis labels for readability
plt.legend()
plt.savefig('Actual_vs_Predicted_W_2022-23.png')
plt.show()
We evaluate the model on the 2022-23 NBA season by calculating the R-squared, MAE, MSE, RMSE, and cross-validation scores.
# Evaluate the model on the 2022-23 season (2nd evaluation)
test_r2 = model.score(X, data_test['W'])
mae = metrics.mean_absolute_error(data_test['W'], y_pred)
mse = metrics.mean_squared_error(data_test['W'], y_pred)
rmse = np.sqrt(mse)
print("\nR-squared (test set):", test_r2)
print("Mean Absolute Error (MAE):", mae)
print("Mean Squared Error (MSE):", mse)
print("Root Mean Squared Error (RMSE):", rmse)
cv_scores = cross_val_score(model, X, data_test['W'], cv=5)
print("Cross-validation scores:", cv_scores)
print("Average cross-validation score:", cv_scores.mean())The output of the above code is shown below:
R-squared (test set): 0.8920766207650307
Mean Absolute Error (MAE): 2.5908684758243763
Mean Squared Error (MSE): 10.092487846927307
Root Mean Squared Error (RMSE): 3.1768676155809996
Cross-validation scores: [0.84426302 0.9368005 0.395935 0.91872027 0.81259453]
Average cross-validation score: 0.7816626626001728
This project is licensed under the MIT License. See the LICENSE file for details.