Model Card: LSTM-based Bitcoin Price Prediction Model

Model Description

This LSTM model is designed to predict the price of Bitcoin (BTC-USD) using historical price data and technical indicators. The model architecture is based on Long Short-Term Memory (LSTM) layers, which are effective for capturing temporal patterns in time series data.

Key Features and Components

Data Preparation
- Data Source: BTC historical data (close, volume, high, and low prices) sourced from Yahoo Finance.
- Technical Indicators: Includes Moving Averages (7-day and 30-day), RSI, and MACD to capture additional insights into price trends and momentum.
- Sequence Creation: Data sequences of 60 days are generated, allowing the model to learn from recent price patterns.
Model Architecture
- LSTM Layers: Two LSTM layers (each with 100 units) to capture sequential dependencies.
- Dropout Layers: Dropout regularization at 20% to prevent overfitting.
- Dense Layers: Two dense layers to output a single price prediction.
Training Configuration
- Loss Function: Huber loss, chosen for its robustness against outliers, making it suitable for financial data which can be volatile.
- Optimizer: Adam optimizer with a learning rate of 0.001 for efficient weight adjustments.
- Early Stopping: Monitors validation loss with a patience of 10 epochs to avoid overfitting.

Performance Metrics

Evaluation Metric: Root Mean Squared Error (RMSE) on both training and test sets to measure prediction accuracy.
Results:
- Training RMSE: X.XX
- Test RMSE: X.XX

Example Use Case

This model can predict Bitcoin's price movement, providing value for day traders or investors looking for short-term trends. It also serves as a foundation for integration into larger financial forecasting systems.

Limitations

Data Sensitivity: The model’s performance heavily depends on the quality and stability of input data.
Overfitting Risks: Despite dropout and early stopping, the volatile nature of price patterns can still lead to overfitting.

Future Improvements

Additional Indicators: Include more advanced technical indicators for better context.
Hyperparameter Tuning: Experiment with different LSTM units, dropout rates, and learning rates to optimize performance.