nba-prediction

Machine learning model to predict NBA games trained on scraped data.

Features

• Game Winner Prediction: Predicts which team will win NBA games
• Comprehensive Statistics: Uses extensive team statistics including traditional and advanced metrics
• Web Scraping Pipeline: Automated data collection using Playwright and BeautifulSoup
• Ridge Classifier Model: Robust machine learning approach for binary classification
• Historical Data: Training on NBA game data with team performance metrics

Technology Stack

• Python: Core programming language
• scikit-learn: Ridge Classifier implementation and model evaluation
• pandas: Data manipulation and analysis
• numpy: Numerical computing
• BeautifulSoup (bs4): HTML parsing for web scraping
• Playwright: Web browser automation for data collection

Model Performance

• Accuracy: 63% on test data
• Algorithm: Ridge Classifier
• Prediction Type: Binary classification (Win/Loss)

Dataset Features

The model uses a comprehensive set of team statistics including:

Traditional Stats: Minutes played (mp), field goals (fg, fga, fg%), 3-pointers (3p, 3pa, 3p%), free throws (ft, fta, ft%), rebounds (orb, drb, trb), assists (ast), steals (stl), blocks (blk), turnovers (tov), personal fouls (pf), points (pts), plus/minus (+/-)

Advanced Stats: True shooting percentage (ts%), effective field goal percentage (efg%), 3-point attempt rate (3par), free throw rate (ftr), rebound percentages (orb%, drb%, trb%), assist percentage (ast%), steal percentage (stl%), block percentage (blk%), turnover percentage (tov%), usage percentage (usg%), offensive rating (ortg), defensive rating (drtg)

Team Context: Home/away status, season information, game dates

Opponent Stats: Complete mirror of all above statistics for opposing teams

Installation

Prerequisites

• Python 3.7 or higher
• scikit-learn
• pandas
• numpy
• bs4
• playwright

Setup

1. Clone the repository:

git clone https://github.com/mlicamele/nba-prediction.git
cd nba-prediction

2. Install required packages:

pip install scikit-learn pandas numpy beautifulsoup4 playwright

3. Install Playwright browsers:

playwright install

Project Structure

nba-prediction/
├── data/
│   ├── scores/           # Game score data
│   └── standings/        # Team standings data
├── get_data.ipynb        # Data collection notebook
├── parse_data.ipynb      # Data processing notebook
├── predict.ipynb         # Model training and prediction notebook
├── nba_games.csv         # Processed game data
├── .gitattributes        # Git LFS configuration
├── .gitignore           # Git ignore file
└── README.md

Usage

The project is organized as Jupyter notebooks for easy exploration and execution:

1. Data Collection

Open and run get_data.ipynb to scrape the latest NBA data:

jupyter notebook get_data.ipynb

2. Data Processing

Process and clean the scraped data using parse_data.ipynb:

jupyter notebook parse_data.ipynb

3. Model Training and Prediction

Train the Ridge Classifier and make predictions with predict.ipynb:

jupyter notebook predict.ipynb

Data Collection

The project uses web scraping to collect NBA statistics:

• Playwright: Handles dynamic web content and browser automation
• BeautifulSoup: Parses HTML content to extract statistical data
• Data Sources: NBA team statistics, game scores, and standings

Model Details

Ridge Classifier

• Type: Linear classifier with L2 regularization
• Advantages: Handles multicollinearity well, robust to overfitting
• Input: Team statistics for both teams in a matchup
• Output: Binary prediction (0 = away team wins, 1 = home team wins)

Feature Engineering

The model incorporates:

• Team performance statistics
• Maximum player statistics per game
• Opponent statistics (complete statistical profile of opposing team)
• Home/away indicators
• Temporal features (season, date)

File Descriptions

• get_data.ipynb: Web scraping implementation using Playwright and BeautifulSoup
• parse_data.ipynb: Data cleaning, processing, and feature engineering
• predict.ipynb: Model training, evaluation, and prediction generation
• nba_games.csv: Final processed dataset with all features
• data/scores/: Raw game score data from scraping
• data/standings/: Team standings and ranking data