Deep-Space CryptoPredictor

A machine learning-based cryptocurrency prediction tool that utilizes real-time crypto data to generate price insights and visualizations.

🚀 Features

• Offline mode to skip network calls (use CSV data only)
• Time-series enrichment: moving averages (5/20), RSI, recent price returns
• Volatility features: rolling volatility at multiple scales (7/14/21 days) and per-symbol z-scores
• Volatility Index: model‑informed, percentile-based index (Low / Medium / High)
• Permutation Feature Importance (PFI) support for explainability
• Visualizations: price charts, prediction vs actual, volatility series and volatility index charts

🛠️ Technologies Used

• CoinGecko API: Primary data source for Crypto prices
• Binance API: Secondary data source
• CSVHelper: For fallback data handling from files

📋 Prerequisites

• .NET 8.0 SDK or newer
• Visual Studio 2022 or VS Code with C# extension

🔧 Setup

1. Clone this repository
2. Open the solution in Visual Studio or VS Code
3. Restore NuGet packages
4. Build and run the application dotnet run -c Release

To run without hitting APIs (use CSVs only), set:

PowerShell

$env:DEEP_OFFLINE="1"
dotnet run -c Release

Outputs are written to the output/ directory with timestamped filenames.

https://github.com/mars-rift/deep-space.git
cd deep-space
dotnet restore
dotnet build
dotnet run

📊 How It Works

1. Data Collection: The application connects to cryptocurrency APIs to fetch Ethereum market data

   • First attempts to use CoinGecko (more reliable in most regions)
   • Falls back to Binance if needed
   • Can use local CSV files as a final fallback

2. Data Preprocessing:

   • Removes outliers and invalid values
   • Applies transformations (log, normalization)
   • Generates technical indicators (moving averages, etc.)

3. Model Training:

   • Trains an ML.NET regression model
   • Uses feature engineering to improve predictions
   • Implements cross-validation and evaluation metrics

4. Visualization:

   • Creates price charts with indicators
   • Displays prediction vs. actual comparison charts
   • Visualizes model residuals for error analysis

🔄 API Usage

CoinGecko API

The application uses CoinGecko's free API to fetch OHLCV (Open-High-Low-Close-Volume) data:

var url = $"{BaseUrl}/coins/ethereum/market_chart?vs_currency=usd&days={days}&interval=daily";

Binance API

As a fallback, the application can use Binance's API:

var url = $"{BaseUrl}/klines?symbol=ETHUSDT&interval={interval}&limit={limit}";

📈 Example Output

When run successfully, the application will:

1. Download the latest Ethereum price data
2. Generate technical indicators
3. Create a price chart in the output directory
4. Train a price prediction model
5. Display prediction accuracy metrics
6. Create a prediction vs. actual visualization
7. Provide a sample prediction for a given Ethereum price

🤝 Contributing

Contributions are welcome! Here's how you can help:

• Add support for additional cryptocurrencies
• Implement more technical indicators
• Improve model accuracy
• Add more visualization options

📄 License

MIT License

---

Note: This project is for educational purposes only. Cryptocurrency investments carry high risk, and predictions should not be used as financial advice.