NBA Game Predictor

A machine-learning model for predicting NBA game outcomes, developed to explore data processing, feature engineering, and building ML pipelines for real-world applications.

Overview

This project uses data from over 10,000 historical NBA games to predict game outcomes with over 60% accuracy. Key features of the project include:

• Data scraping from various NBA statistics sources using Python, Playwright, and BeautifulSoup
• Data organization and enhancement with Pandas
• Feature selection using Sequential Feature Selection to simplify the dataset
• Prediction model using Ridge Regression and backtesting for validation

Getting Started

1. Clone the Repository

   git clone https://github.com/yourusername/NBA-Game-Analytics-Predictor.git
   cd NBA-Game-Analytics-Predictor
   

To run the data scraper for schedule and games, execute:

python get_data.py

Data Preparation

After collecting raw data, run the data preparation script to clean, organize, and enhance it with additional statistics:

python parse_data.py

This step produces a processed dataset ready for feature selection and model training.

Machine Learning Pipeline

Feature Selection: Using Sequential Feature Selection, the dataset is optimized to use only the most predictive features.

Model Training and Backtesting: Ridge regression is used to train the model, with backtesting applied for performance evaluation.

To train the model and test its accuracy:

• Run predict.ipynb