BTC-RNN-LSTM-GRU-Forecasting

Overview

This project focuses on predicting Bitcoin daily log returns using three neural network architectures: RNN, LSTM, and GRU. The repository includes implementations of each model and uses a dataset with market, macroeconomic, and crypto-specific indicators.

Project Contents

• BTC_RNN.ipynb: RNN model for log return forecasting.
• BTC_LSTM.ipynb: LSTM model for handling long-term dependencies.
• BTC_GRU.ipynb: GRU model for efficient sequence modeling.
• Open-Open-Dataset-BTC.csv: Dataset used in the project.
• README.md: Project documentation.

Dataset

The dataset contains:

• Market data: Open, High, Low, Close, Volume, MarketCap.
• Crypto-specific metrics: Fear & Greed Index, CBBI, VDD, Hash Rate, Days Since Halving.
• Macroeconomic indicators: Inflation, M2SL, S&P500, DXY, VIX of VIX.
• Moving Averages: MA5, MA20.

Methodology

1. Preprocessing: Data normalization, lagged feature creation.
2. Models:
   • RNN: Baseline sequence model.
   • LSTM: Handles long-term dependencies.
   • GRU: Simplified alternative to LSTM.
3. Evaluation: Metrics include MAE, RMSE, and MAPE.