TKO Analytics

An advanced UFC analytics platform powered by AI that provides deep insights into fight statistics, predictions, and performance metrics.

Demo

Click the image above to watch the demo video!

Installation & Usage

Prerequisites

• Python 3.10
• Conda package manager
• OpenAI API key

Setup Instructions

1. Create a new Python 3.10 environment using Conda:

conda create -n tko-analytics python=3.10
conda activate tko-analytics

2. Clone the repository:

git clone https://github.com/jaysomp/tko-analytics
cd tko-analytics

3. Install required dependencies:

pip install -r requirements.txt

4. Configure OpenAI API key:

   • Create a .env file in the project root directory
   • Add your OpenAI API key:

   OPENAI_API_KEY="your-api-key-here"
   

5. Run the application:

streamlit run app.py

Features

Core Features

• Natural language interaction with UFC fight data
• Real-time statistical analysis and visualization
• Machine learning-powered fight predictions
• Multi-dimensional data exploration
• Automated insight generation
• Interactive data visualization

Data Analysis

• Comprehensive UFC fight database updated through November 2024
• Multi-table query support for complex statistical analysis
• Real-time data visualization using PandasAI
• Advanced fight outcome predictions using XGBoost ML models
• Historical fight statistics and performance metrics
• Betting odds analysis and probability calculations

Interactive Interface

• Natural language query processing using GPT-4
• Conversational AI assistant for data exploration
• Dynamic visualization generation
• Interactive dashboard with configurable views
• Multi-table data exploration capabilities
• Context-aware follow-up questions
• Custom table selection interface
• Real-time query modification
• Flexible window configuration
• Historical conversation tracking
• Integrated visualization panels

Analysis Capabilities

• Fight outcome predictions with confidence scores
• Fighter performance trend analysis
• Betting odds analysis and probability calculations
• Head-to-head fighter comparisons
• Fighting style analysis
• Win/loss pattern recognition
• Title fight statistics
• Performance metrics tracking
• Historical matchup analysis
• Round-by-round statistics

Technical Capabilities

• SQL query optimization for complex multi-table analysis
• BCNF-normalized database structure
• Real-time data processing and visualization
• Machine learning integration for predictive analytics
• Scalable architecture supporting live data updates
• Helper functions for optimized LLM context management
• Hybrid visualization approach for both single and multi-table analysis
• Token-optimized database schema handling

Application Design

Architecture Overview

graph TD
    A[XgBoost Model] --> B[Results]
    B --> D[Predictions]
    E[SQLAgent] <-.-> F[Database]
    G[Tools] --> E
    G --> H[Pandasai]
    H <-.-> F
    I[LLM GPT-4] <-.-> G
    J[Helper Functions] --> I
    D <-.-> G
    J <-.-> F

Loading

Key Components

1. Core Analysis Engine

• SQL Query Tool: Handles complex multi-table queries using LangChain's SQL agent
• Visualization Engine: PandasAI integration for dynamic chart generation
• Prediction System: XGBoost-based model for fight outcome predictions
• Data Pipeline: Optimized for real-time updates and analysis

2. LLM Integration

• Natural language query parsing
• Context-aware conversation management
• Automated query optimization
• Dynamic response generation

3. Optimization Features

• Smart caching for frequent queries
• Token-optimized database schema handling
• Efficient multi-table join operations
• Memory-optimized result storage

4. User Interface Components

• Interactive chat interface
• Dynamic visualization panel
• Table selection sidebar
• Configurable analysis windows
• Conversation history tracking

Database Schema

Core Tables

• Fighters: FighterID (PK), Name
• Fights: FightID (PK), Date, Location, Winner, FinishDetails, FinishRound, FinishRoundTime, TotalFightTimeSecs, RedFighterID (FK), BlueFighterID (FK)
• Odds: OddsID (PK), FightID (FK), RedOdds, BlueOdds, RedExpectedValue, BlueExpectedValue, RedDecOdds, BlueDecOdds, RSubOdds, BSubOdds, RKOOdds, BKOOdds

Technology Stack

• Backend: Python, SQLite
• ML/AI: XGBoost, GPT-4, LangChain
• Data Analysis: PandasAI
• Visualization: Custom plotting tools with multi-table support

Future Development

• Real-time fight analytics pipeline
• Enhanced prediction models incorporating fighting styles
• Live prediction capabilities
• Interactive visualization improvements
• React-based frontend migration
• Personalized user dashboards

Authors

• Jaydeep Sompalli
• Jeet Arora