---
tags:
- time-series
- forecasting
- cryptocurrency
- LSTM
- deep-learning
datasets:
- HistoricalCryptoData
license: mit
---

# 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. 

1.  **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).
2.  **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.
3.  **Dense Output Layer:** A final fully connected layer projects the output of the LSTM layers to the desired **1-day ahead closing price prediction**.
4.  **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.