A TypeScript implementation of a neural network for cryptocurrency price prediction using real-time market data from Binance. The system uses a 2-2-1 neural network architecture with sigmoid activation functions to predict future price movements based on historical OHLCV data.

• 📊 Real Market Data: Live BTC/USDT data from Binance exchange via CCXT
• 🧠 Neural Network: 2-2-1 architecture with sigmoid activation and backpropagation
• 🔮 Price Prediction: Predicts next price based on historical close prices
• 📈 Confidence Scoring: Calculates prediction confidence based on recent accuracy
• ⚡ TypeScript: Fully typed implementation with strict TypeScript configuration

# Install dependencies
npm install

# Development mode
npm run dev

# Build and run
npm run build
npm start

import { MarketPredictor, MarketDataProvider } from '@/core'

// Configure neural network
const config = {
  epochs: 60,
  learningRate: 0.1,
  useSimpleBackprop: false
}

// Create predictor and data provider
const predictor = new MarketPredictor(config)
const dataProvider = new MarketDataProvider()

// Fetch real market data
const marketData = await dataProvider.fetch4H('BTC/USDT', 100)

// Add data and train
marketData.forEach(data => predictor.addMarketData(data))
const lossCurve = predictor.train()

// Make prediction
const prediction = predictor.predict()
console.log(`Predicted: $${prediction.predictedOutput}`)
console.log(`Confidence: ${prediction.confidence * 100}%`)

• 📥 Input Layer: 2 nodes (previous close prices)
• 🔗 Hidden Layer: 2 nodes with sigmoid activation
• 📤 Output Layer: 1 node (predicted price)
• 🎯 Training: Backpropagation with MSE loss

• 📡 MarketDataProvider: Fetches real-time data from Binance
• 🎯 MarketPredictor: High-level prediction interface
• 🧠 NeuralNetwork: Core neural network implementation

• 🏦 Exchange: Binance (futures)
• 💰 Symbol: BTC/USDT
• ⏰ Timeframes: 1H, 4H, 1D
• 📊 Data: OHLCV (Open, High, Low, Close, Volume)

# Code quality
npm run lint
npm run format

# Type checking
npm run build

# Testing
npm test

src/
├── core/
│   ├── MarketDataProvider.ts    # Real-time data fetching
│   ├── MarketPredictor.ts       # High-level prediction interface
│   └── NeuralNetwork.ts         # Core neural network
├── types/
│   └── index.ts                 # TypeScript type definitions
└── index.ts                     # Main entry point

The neural network implements the following logic:

1. Data Normalization: Scale prices to [0,1] range using max price scaling
2. Feature Engineering: Use previous two close prices as input features
3. Training: Backpropagation with gradient descent and MSE loss function
4. Prediction: Sigmoid activation for price forecasting with denormalization
5. Confidence: Error-based confidence calculation using recent prediction accuracy
6. Weight Initialization: Seeded random weights for reproducible results

• Training Time: ~60 epochs per prediction with configurable learning rate
• Data Points: 100-500 historical OHLCV prices from Binance
• Accuracy: Varies based on market volatility and trend conditions
• Confidence: 0-100% based on recent prediction accuracy
• Memory Usage: Efficient with minimal memory footprint
• Network State: Saveable/loadable neural network weights and configuration

interface NeuralNetworkConfig {
  epochs: number        // Training iterations (10-1000)
  learningRate: number  // Gradient descent rate (0.00001-1.0)
  useSimpleBackprop: boolean // Backpropagation algorithm
}

MIT License - see LICENSE file for details