Cricket Match Prediction API

This repository contains a machine learning model that predicts cricket match outcomes based on various in-game statistics.

Project Structure

cricket/
├── data/
│   ├── train/
│   │   └── cricket_dataset.csv
│   ├── test/
│   │   └── cricket_dataset_test.csv
│   └── results/
│       └── prediction_results_{timestamp}.csv
├── src/
│   ├── app.py
│   └── eval_api.py
│    ├── model/
│       └── cricket_prediction_model.pkl
│    ├── notebooks/
│       └── cricket_prediction.ipynb
└── README.md

Dataset Features

The model uses the following features to make predictions:

• total_runs: Total runs scored
• wickets: Number of wickets fallen
• target: Target score
• balls_left: Number of balls remaining
• won: Target variable (1 for win, 0 for loss)

Model Selection

Random Forest Classifier was chosen for this task because:

1. It handles non-linear relationships effectively
2. Provides good protection against overfitting
3. Can capture complex interactions between features
4. Works well with both numerical and categorical data

Setup and Installation

1. Clone the repository:

git clone https://github.com/zee-rox/cricket.git
cd cricket

2. Install required dependencies:

pip install requirements.txt

Running the Server

1. Start the FastAPI server:

cd src
uvicorn app:app --reload

The server will start at http://localhost:8000

API Endpoints

POST /predict

Makes predictions on cricket match outcomes using a CSV file.

Request body:

{
    "csv_path": "path/to/input.csv"
}

Response:

{
    "result_path": "path/to/results/prediction_results_{timestamp}.csv"
}

Evaluating the API

1. Run the evaluation script:

python src/eval_api.py

This will:

• Send a test CSV to the API
• Print the API response
• Display prediction results
• Clean up test files

Model Performance

The Random Forest model achieves:

• High accuracy in predicting match outcomes
• Good balance between precision and recall
• Robust performance across different match scenarios

Error Handling

The API includes comprehensive error handling for:

• Missing CSV files
• Invalid data formats
• Empty datasets
• Server errors

Results

Predictions are saved in the data/results directory with timestamps for easy tracking and analysis.

Notes

• The model filters matches where:
  • Less than 60 balls are remaining
  • Target score is greater than 120
• Predictions are binary (win/loss)
• Results include all original features plus predictions