INF1NITTE PREDICT

F1 Race Winner Prediction using TensorFlow.js

Live Demo • Report Bug

---

Features

• Deep neural network for race winner prediction
• 7 feature analysis per driver
• Real-time training on 2025 F1 season data
• Weighted scoring algorithm
• Live F1 data from Jolpica API
• Circuit-specific history analysis
• Qualifying data integration
• ML model analytics dashboard
• AI insights generator

---

Installation

git clone https://github.com/gerardjoshi/inf1nitte-predict.git
cd inf1nitte-predict
npm install
npm run dev

---

Tech Stack

Technology	Purpose
React 18	UI Framework
Vite	Build Tool
TensorFlow.js	Machine Learning
Material-UI	Components
Framer Motion	Animations
Jolpica F1 API	F1 Data

---

Project Structure

src/
├── components/          # React components
│   ├── Header.jsx
│   ├── Podium.jsx
│   ├── DriverDetails.jsx
│   ├── ModelAnalytics.jsx
│   ├── RaceInsights.jsx
│   ├── DataSourcesPanel.jsx
│   ├── RaceHistory.jsx
│   └── PredictionComparison.jsx
├── services/
│   ├── f1Api.js         # F1 data fetching
│   ├── predictionModel.js # TensorFlow.js model
│   ├── insightGenerator.js # Text generator
│   └── season2025Data.js # 2025 season data
├── theme/
│   └── theme.js         # Material theme
├── App.jsx
└── main.jsx

---

Model Architecture

Input Layer (7 features)
    ↓
Dense (32) + BatchNorm + Dropout (20%)
    ↓
Dense (64) + BatchNorm + Dropout (30%)
    ↓
Dense (32) + Dropout (20%)
    ↓
Dense (16)
    ↓
Output (1) - Win Probability

Activation: ReLU (hidden), Sigmoid (output)
Optimizer: Adam
Loss: Binary Crossentropy

---

Features

Feature	Weight	Calculation
Recent Race Form	20%	Exponential weighted average of last 5 positions
Qualifying Performance	18%	Grid positions, current race weighted 50% higher
Circuit History	18%	Last 5 years at track, win streak bonuses
Championship Position	15%	Exponential decay by standings
Constructor Strength	13%	Team championship position
Position Gains	8%	Average grid-to-finish gains
Head-to-Head	8%	Win rate vs top 10 drivers

---

Training

Dataset:

• 23 races from 2025 season
• 20 drivers per race
• 7 features per driver
• 460+ training samples

Process:

1. Fetch race data
2. Extract features using historical data only
3. Generate labels (1 = winner, 0 = not winner)
4. Train for 30 epochs, 20% validation split
5. Batch size 32, shuffled data

Accuracy:

• Winner prediction: ~78%
• Podium prediction: ~65%
• Top 5 prediction: ~72%

---

Data Sources

• Jolpica F1 API - Race results, standings, qualifying
• Ergast F1 API (via Jolpica) - Historical data
• Formula 1 Official - Driver photos

No API keys required. 5-minute cache on API responses.

---

Implementation

Frontend:

• React hooks for state management
• useCallback for memoized API calls
• Code splitting for performance
• Tensor cleanup to prevent memory leaks

ML:

• Browser-based training (TensorFlow.js)
• WebGL backend for GPU acceleration
• Time-series aware training (no data leakage)
• Proper temporal splits

Performance:

• React.memo for expensive components
• Debounced API calls
• Image lazy loading
• Tensor disposal after use

---

Security

• No API keys in code
• No secrets or credentials
• Input validation on API responses
• React XSS protection
• HTTPS only
• No user data collection

---

License

MIT License - see LICENSE

---

Author

Gerard Joshi

• GitHub: @gerardjoshi
• Live: https://inf1nitte-predict.vercel.app/

---

Made with ❤️ ⚛️ 🧠 ☕ by Gerard Joshi