MLB Elo Ratings

Overview

This project provides tools and notebooks for recreating and building on FiveThirtyEight's Elo Rating system for baseball teams. It includes an EloTracker class to automatically produce the latest Elos for each team, given the results presented in an inputted box score dataframe. It also includes an MLBSimulator class to simulated a given schedule dataframe, and see which teams are most successful, using Elos to estimate per-game win probabilities.

---

Project Highlights

• Improved Elo rating system: Re-implemented and enhanced FiveThirtyEight’s Elo model for MLB teams by learning custom rating adjustments for per-game variables, including home advantage, rest days, travel distance, and pitcher ability via logistic regression. This improved 2025 season prediction accuracy from 55.6% to 56.2%.
• Season simulations: Leveraged custom Elo ratings to simulate full seasons, generating expected wins and playoff probabilities, improving $R^2$ for 2025 regular season wins from 0.46 to 0.49.
• Exploring rule changes: Currently experimenting with simulating seasons under potential rule changes to understand team impacts.

---

Project Structure

analysis/
├── **data/**  
│   ├── **clean/** – Preprocessed and cleaned datasets  
│   └── **raw/** – Original raw CSV files  
├── src/
│   ├── preprocess/
│   │   └── notebooks/            # Notebooks for cleaning and preprocessing data
│   ├── simulation/
│   │   ├── simulation.py         # Includes MLBSimulator class
│   │   └── __init__.py
│   ├── utils/
│   │   ├── math_utils.py         # Math utility functions
│   │   ├── misc_utils.py         # Miscellaneous utility functions
│   │   └── __init__.py
│   └── elos/
│       ├── elo_tracker.py        # Includes EloTracker class
│       └── __init__.py
└── notebooks/
    ├── feature_analysis/         # Analysis of team performance and features
    ├── simulation/               # Running simulation notebooks
    ├── model_selection/          # Comparing models and evaluation vs 538 baseline
    └── presentation/             # Presentation-ready notebooks and visualizations

---

Installation

1. Clone the repository:

git clone https://github.com/lanceHEN/Elo-Ratings.git
cd analysis

2. Create a virtual environment:

python -m venv venv
source venv/bin/activate   # Mac/Linux
venv\Scripts\activate      # Windows

3. Install dependencies:

pip install -r requirements.txt

Usage

Preprocessing Data

Notebooks under src/preprocess/notebooks/ are for cleaning raw CSVs and generating cleaned datasets in data/clean/.

Simulations

Use scripts and notebooks under src/simulation/ and src/notebooks/simulation/ to run Elo simulations and other analyses.

Feature Analysis and Model Selection

notebooks/feature_analysis/ and notebooks/model_selection/ contain notebooks for exploring features, comparing models, and evaluating predictions.

Utilities

src/utils/ contains helper modules:

• math_utils.py — mathematical and statistical functions
• misc_utils.py — miscellaneous helper functions

src/elos/elo_tracker.py provides the EloTracker class for tracking and updating Elo ratings.

Data

data/raw/ — Original datasets

data/clean/ — Cleaned datasets generated from preprocessing notebooks

Recommended workflow: preprocess raw CSVs -> generate cleaned datasets -> run simulations and analyses