CryptocurrencyPriceForecaster
Overview
This model is a deep learning-based Long Short-Term Memory (LSTM) network designed for multivariate time series forecasting of cryptocurrency prices (specifically, the next 7-day closing price of BTC/USD). It utilizes historical price data and technical indicators as input features.
Model Architecture
The architecture is a classic sequence-to-sequence structure implemented with LSTMs, optimized for handling temporal dependencies.
- Input Layer: Takes a sequence of the last 60 time steps (e.g., 60 days) of 7 features (Open, High, Low, Close, Volume, Moving_Avg_14, RSI_14).
- LSTM Layers: 3 stacked LSTM layers with a hidden size of 256 neurons, incorporating dropout (0.2) to prevent overfitting. LSTMs are crucial for capturing long-term dependencies in price movements.
- Dense Output Layer: A final fully connected layer projects the output of the LSTM layers to the desired 1-day ahead closing price prediction.
- Training: Trained using the Mean Squared Error (MSE) loss function.
Intended Use
This model is strictly for research and experimental financial modeling.
- Prediction: Forecasting the next day's closing price for BTC/USD based on a 60-day window.
- Feature Importance Analysis: Studying the predictive power of different technical indicators (RSI, Moving Averages).
- Simulated Trading: Use in a paper trading environment to test the viability of model-driven trading signals.
Limitations
- Not Financial Advice: This model is a statistical tool and its predictions should NOT be used as the sole basis for real financial investment decisions. Cryptocurrency markets are highly volatile.
- Exogenous Factors: The model only uses technical (price/volume) data. It does not account for sudden, unpredictable external events (e.g., regulatory changes, major exchange hacks, macroeconomic shocks) which are often primary drivers of crypto volatility.
- Prediction Horizon: Optimized for short-term (1-day) forecasting. Accuracy degrades substantially beyond this horizon.
- Stationarity: The input data requires careful normalization (
MinMaxScaler used) and handling of non-stationarity to ensure reliable training.
