●5 reads●MIT License
Crypto Predictor | LSTM Model | Streamlit
Table of contents
🪙 CryptoPredictor
Predicting Cryptocurrency Prices with ML
CryptoPredictor is a machine learning-powered app designed to forecast cryptocurrency prices. Built with Python and Streamlit, it utilises historical data to predict trends for popular cryptocurrencies, empowering users with data-driven insights.
🧬 Project Structure
CryptoPredictor ├── assets/ │ ├── dataCleaning/ │ │ ├── CoinGeckoToCsv.ipynb │ │ └── cryptoTickers.csv │ └── gifs/ │ └── crypto.gif ├── streamlit_app/ │ ├── modules/ │ │ └── helper.py │ ├── pages/ │ │ └── 01_₿__CryptoPredictor.py │ └── 00_🛈_Info.py ├── LICENSE ├── README.md └── requirements.txt
🛠️ How It's Built
CryptoPredictor is built with the following core frameworks and tools:
- Streamlit - To create an intuitive web interface.
- Yahoo Finance API (YFinance) - To fetch up-to-date cryptocurrency data.
- CoinGecko API - To dynamically fetch the top 2,500 cryptocurrencies by market capitalisation from a pool of 17,061 available tokens, ensuring access to real-time data for popular selections. For an updated list of the 2,500 tickers based on market capitalisation, users can run the Jupyter Notebook available at: CoinGeckoToCsv.ipynb.
- LSTM (Long Short-Term Memory) - A neural network model optimised for time-series predictions.
- Plotly - To generate dynamic and interactive financial charts.
- Pandas - To manipulate and process cryptocurrency datasets.
🧑💻 How It Works
- The user selects a cryptocurrency (e.g., BTC, ETH).
- Historical price data is retrieved using the Yahoo Finance API.
- The LSTM model is trained on the past 60 days of historical data.
- Predictions are generated for the next 1–90 days.
- Results are displayed with interactive charts and tables.
✨ Key Features
- Real-time cryptocurrency data - Access accurate and up-to-date information.
- Interactive charts - View historical trends and future predictions visually.
- Custom prediction ranges - Forecast prices for 1 to 90 days.
- Downloadable CSV - Save prediction results for further analysis.
- User-friendly interface - Accessible for novice and experienced users alike.
🚀 Getting Started
Local Installation
- Clone the repository:
git clone https://github.com/user/CryptoPredictor.git
Hint: Replace user with josericodata in the URL above. I am deliberately asking you to pause here so you can support my work. If you appreciate it, please consider giving the repository a star or forking it. Your support means a lot—thank you! 😊
- Navigate to the repository directory:
cd CryptoPredictor
- Create a virtual environment:
python3 -m venv venvCrypto
- Activate the virtual environment:
source venvCrypto/bin/activate
- Install requirements:
pip install -r requirements.txt
- Navigate to the app directory:
cd streamlit_app
- Run the app:
streamlit run 00_🛈_Info.py
The app will be live at http://localhost:8501
🎬 Demo
Crypto Predictor Page:
▶️ Watch the YouTube Tutorial
Click the image above or here to watch the video on YouTube.
🔮 Future Enhancements
Planned improvements and new features include:
- Integration of advanced ML models (e.g., Prophet) for better prediction accuracy.
- Support for additional cryptocurrencies to expand coverage.
- Volatility analysis to measure price swings and potential risks.
- User accounts and history tracking for tailored predictions and personalised experiences.
🔧 Environment Setup
The CryptoPredictor app is built and tested using the following software environment:
- Operating System: Ubuntu 22.04.5 LTS (Jammy)
- Python Version: Python 3.10.12
Ensure your environment matches or exceeds these versions for optimal performance.
📋 Important Notes
- CoinGecko API Error 429: If too many requests are made to the URL, the API may block further requests. Please restart or close the app, and try again after a minute or two. The data should then be available.
- Using the Crypto Predictor:
- Select a cryptocurrency from the dropdown menu.
- Choose the desired prediction range using the slider.
- Adjust the Epochs slider to set the number of training iterations. Note: Higher epochs result in longer training times but can improve model accuracy.
- Click the Run Prediction button to generate results.
🤝 Open Pull Requests
If you find any bug, feel free to contact me by opening a pull request on GitHub or via email at maninastre@gmail.com.
⚠️ Disclaimer
This app is designed to demonstrate my skills in data modeling and analytics, showcasing how data-driven insights can assist in building my portfolio as a data analyst. It is not intended to provide financial advice or investment guidance. The predictions are for illustrative purposes only and should not be relied upon for making financial decisions.