swapnil-ahlawat / NBA_Game_Predictor

A comparative study of various models for prediction of Win/Loss of a basketball game based on the team’s as well as players’ past statistics. Also focused on the web scraping techniques to scrap raw datasets from the nba/stats website and feature engineering on the collected datasets to best suit the classification problem.

Prediction of NBA Games Using Machine Learning

Folder Contents:

'data' Folder:

File	Content
PlayerScheduleXX-XX.csv	Raw Player Box Scores
schedule.csv	Raw Team Box Scores
NBAGameDataset.csv	NBA Game Features Dataset

'src' Folder:-

File	Content
webScraping.py	Web Scraping script
featureEngineering.py	Raw Data to Features Script
player.py	Model training and testing on Players Statistics
team.py	Model training and testing on Team Statistics
teamAndPlayer.py	Model training and testing on Combined Team and Players Statistics

Note:- All the required libraries are mentioned in 'requirements.txt' and can be directly pip installed.

Web Scraping

Run python script 'webscraping.py' to scrap raw data from websites. This will generate a lot of csv files of raw data. This will take a lot of time as the pages are dynamically loaded and javascript clicking, and others are done by script. All the csv files are already in the data folder.

>> python3 ./src/webscraping.py

Feature Engineering

Run python script 'featureEngineering.py' to generate features from raw data. This will take around 10 minutes to run. The final csv dataset is already in the data folder with the name "NBAGameDataset.csv".

>> python3 ./src/featureEngineering.py

Running Models:-

Hyperparamter tuning code is commented as it takes hours to run. The models with tuned hyperparameters is added at the end of the scipt so that you can get the results faster. The models will load the csv from data folder.

To see results when models were feeded only team features:- run python script 'team.py'

>> python3 team.py

To see results when models were feeded only player features:- run python script 'player.py'

>> python3 player.py

To see results when models were feeded both team and player features:- run python script 'teamAndPlayer.py'

>> python3 teamAndPlayer.py

Note: The coorelation matrix has been commented out because it gives error on running from command line. In case if you want to try that code on notebook, please have matplotlib library installed.