📊 MLB Wins Predictor

⚾ Overview

The MLB Wins Predictor is a machine learning project that forecasts the number of wins a Major League Baseball (MLB) team will achieve in a season based on key performance metrics. This project utilizes a Keras neural network model trained on historical team data, with a FastAPI backend serving real-time predictions via a RESTful API.

---

🚀 Features

• Machine Learning Model: Predicts team wins using a neural network.
• FastAPI Endpoint: Serves predictions via a /predict endpoint.
• Scalable Deployment: Deployed on an AWS EC2 instance with Uvicorn.
• Model Training Pipeline: Includes feature selection, hyperparameter tuning, and model evaluation.

---

💽 Project Structure

mlb-wins-predictor/
│
├── app.py                # FastAPI application for serving predictions
├── mlb_wins_model.h5     # Trained Keras model (HDF5 format)
├── scaler.pkl            # StandardScaler for data normalization
├── pitching_batting_combined_df.csv  # Combined dataset for model training
├── README.md             # Project documentation
├── requirements.txt      # Python dependencies
└── tuner_results/        # Keras Tuner results for hyperparameter tuning

---

📈 Model Training Workflow

1. Data Preparation:
   • The dataset pitching_batting_combined_df.csv contains various performance metrics (batting and pitching).
   • Features are selected based on their importance to predict wins (W).
2. Feature Selection:
   • Used a Random Forest model to identify the top 10 features contributing to win predictions:
     • WAR, RS, SO, SV, ER, R, SH, GB, ERA, BB
3. Model Training:
   • Neural network with two dense layers, dropout for regularization, and ReLU activations.
   • Hyperparameter tuning with Keras Tuner to optimize model architecture.
4. Model Evaluation:
   • Evaluated on a test set using Mean Absolute Error (MAE) and Mean Squared Error (MSE).

---

🛠️ Tech Stack

• Python (3.10)
• TensorFlow / Keras
• Scikit-Learn
• FastAPI
• Uvicorn
• Joblib
• AWS EC2 (Deployment)

---

⚙️ Setup Instructions

1. Clone the Repository

git clone https://github.com/ChadB12/mlbwinspredictor.git
cd mlbwinspredictor

2. Create a Virtual Environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Ensure Model and Scaler Files Are Present

Place the following files in the root directory:

• mlb_wins_model.h5: Trained Keras model.
• scaler.pkl: StandardScaler for input data normalization.

5. Run the FastAPI App

uvicorn app:app --host 0.0.0.0 --port 8000

The API should now be running at http://0.0.0.0:8000.

---

🔮 Usage

Endpoint: /predict

• Method: POST
• Content-Type: application/json

• Request Format:

  {
    "data": [[45, 850, 1300, 45, 520, 650, 30, 1400, 3.50, 450]]
  }
  

• Example Response:

  {
    "predictions": [90.5]
  }
  

Using curl:

curl -X 'POST' \
  'http://your_server_ip:8000/predict' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
    "data": [[45, 850, 1300, 45, 520, 650, 30, 1400, 3.50, 450]]
  }'

---

📊 Example Data

Typical input values for the model based on an above average MLB team’s stats:

Feature	Description	Example Value
WAR	Wins Above Replacement	45
RS	Runs Scored	850
SO	Strikeouts by pitchers	1300
SV	Saves	45
ER	Earned Runs	520
R	Runs allowed by pitchers	650
SH	Sacrifice Hits	30
GB	Ground Balls	1400
ERA	Earned Run Average	3.50
BB	Walks allowed by pitchers	450

---

👌 Testing the API

1. Go to Docs: Open the Swagger UI to test the API via a user-friendly interface:

   http://18.219.97.159:8000/docs
   

2. Interactive API Documentation: Use the Swagger interface to send POST requests and view responses.

Dataset Used

The dataset used for training the model includes a combination of pitching and batting statistics. The dataset can be found in the following path within the repository:

Dataset

👨‍💻 Author

• Chad Broussard
• LinkedIn: LinkedIn Profile
• GitHub: GitHub Profile

---

🌟 Acknowledgments

• TensorFlow and Keras for model building.
• FastAPI for creating the API.
• AWS EC2 for deployment.
• Scikit-Learn for data preprocessing and feature selection.
• Pybaseball for data accumulation.