A deep learning project using LSTM networks to forecast Bitcoin prices based on historical time series data. Built with TensorFlow, Keras, Pandas, and NumPy.
- Data preprocessing and normalization
- Sequence creation for time series modeling
- LSTM and Bidirectional LSTM architectures
- Model training and evaluation
- Future price prediction with aligned timestamps
- Visualization of predictions vs actual prices
- Python
- TensorFlow / Keras
- Pandas / NumPy
- Matplotlib
git clone https://github.com/lina2016/bitcoin-prediction.git
cd bitcoin-prediction
pip install -r requirements.txt
bitcoin-prediction/
├── data/ # Raw and processed CSV files
├── notebooks/ # Jupyter notebooks for exploration
├── models/ # Saved models and weights
├── utils/ # Helper functions
├── main.py # Training and prediction script
└── README.md # Project overview
## 📂 Notebooks
Open the notebooks in order:
1. `data_processing.ipynb`
2. `model_LSTM.ipynb`
3. `prediction.ipynb`
---
## 📊 Model Overview
- **Architecture**: LSTM with sequence batching and dropout
- **Input**: Normalized Bitcoin price sequences
- **Output**: Predicted future prices with timestamp alignment
- **Evaluation**: Mean Squared Error (MSE), visual comparison with actual prices
---
## 🧪 Sample Output
| Date | Predicted Price |
|------------|-----------------|
| 2025-08-05 | $29,842.17 |
| 2025-08-06 | $30,104.55 |
**Visualizations include:**
- Price trends over time
- Future forecast plot
---
## 📌 Author
**Lina** — AI TensorFlow Developer | Time Series Enthusiast
📍 Gilroy, CA
🔗 [LinkedIn Profile](https://www.linkedin.com/in/lina-jamadar/)
---
## 🧠 Future Work
- Compare LSTM with CNN, RNN, and DNN architectures
- Add hyperparameter tuning and model selection
- Deploy model via Flask or Streamlit for interactive forecasting
---
## 📜 License
This project is open-source and available under the **MIT License**.