UFC Parlay Predictor: An AI-Powered MMA Prediction Platform

UFC Parlay Predictor is a UFC/MMA fight prediction platform that integrates machine learning, dynamic ELO rating system, and real-time fighter data. It aims to provide data-driven fight outcome predictions and sports betting decision support for fight enthusiasts and analysts. The platform uses FastAPI backend, installable PWA frontend, and is deployed on a cloud-native architecture, demonstrating how to transform a data science project into a production-grade application.

• Original Author/Maintainer: biggpower08
• Source Platform: GitHub
• Original Title: UFC-Parlay-Predictor
• Original Link: https://github.com/biggpower08/UFC-Parlay-Predictor
• Release Date: May 26, 2026

This project is specifically designed for MMA (UFC/MMA) event analysis and sports betting decision support. Its core is to provide data-driven prediction services by combining machine learning, dynamic ELO rating system, and real-time fighter data.

Backend Architecture

Built with FastAPI framework, core dependencies include:

• Data Storage: Supabase Postgres
• ORM and Migration: Alembic
• ASGI Server: Uvicorn

Frontend Architecture

Next.js-based PWA with installability, offline capability, and native experience. Configuration files include app/frontend/public/manifest.json, service-worker.js, icon.svg.

Cloud-Native Deployment

• Primary Platform Render: Configured with render.yaml, frontend static app is served by FastAPI backend. Startup command: alembic upgrade head && uvicorn ufc_predictor.api.app:app --host 0.0.0.0 --port $PORT
• Alternative Platform Vercel: Optional deployment plan pointing to Render API
• Data Layer Supabase: Hosts Postgres database and provides connection credentials

Dynamic ELO Rating System

Implements a variant of the chess ELO system, dynamically adjusts fighter ratings based on match results, and compares performance of different version algorithms via the python scripts/compare_elo_versions.py script.

Machine Learning Models

Trains models using fighter historical data, ELO ratings, and match features. Models are stored in pickle format; SQLite cache is used in development environment, while Supabase Postgres is used in production.

Real-Time Data Integration

Imports real-time fighter data via the python scripts/update_live_database.py --apply-schema script.

Database Initialization

1. Create a Supabase project and obtain connection information
2. Set the DATABASE_URL environment variable
3. Run migration: alembic upgrade head
4. Import source data: python scripts/import_supabase.py (Note: Import CSV instead of SQLite cache; do not upload fighters.db or model pickle to Supabase)

Render Environment Variables

Required variables include DATABASE_URL, SUPABASE_URL, SUPABASE_SERVICE_KEY, UFC_PREDICTOR_DATA_DIR, FRONTEND_ORIGINS, LOG_LEVEL

Health Check and Local Development

• Health check endpoints: GET /api/health, GET /health
• Backend startup: python -m uvicorn ufc_predictor.api.app:app --host 127.0.0.1 --port 8000
• Frontend development: cd app/frontend && npm install && npm run dev (proxies to backend by default)

• Data Management: Uses SQLite cache in development environment and Supabase Postgres in production to avoid sensitive data leakage
• Architecture Evolution: Supports algorithm version A/B testing and rollback via compare_elo_versions.py
• Multi-Platform Deployment: Primary Render platform + alternative Vercel platform to improve availability and fault tolerance

Application Scenarios

• Sports betting analysis: Provides betting decision support
• Fight event reporting: Automatically generates pre-match analysis
• Fantasy sports games: Provides rating references
• Academic research: Case study for sports data analysis

Future Expansion

• Introduce features like fighter physical fitness and injury history
• Implement real-time odds comparison
• Add social media sentiment analysis
• Build a fighter similarity recommendation system

UFC Parlay Predictor demonstrates how to engineer a data science project into a production-grade application, covering full-stack development aspects such as machine learning models, API design, PWA experience, and cloud-native deployment. Its clear architectural layers, comprehensive deployment documentation, and multi-platform support strategy provide valuable references for similar projects.