gianlucamazza / btc-prediction-lstm

Bitcoin Price Prediction Using LSTM

Bitcoin Price Prediction Using LSTM

Project Overview

This project uses a Long Short-Term Memory (LSTM) model to predict Bitcoin prices. It encompasses data preparation, feature engineering, model building, training, and predictions using historical price data.

Motivation

The aim is to apply machine learning techniques to financial data, developing a model that accurately predicts Bitcoin prices, potentially aiding in investment decisions.

Structure

• data_preparation.py: Processes raw Bitcoin price data.
• data_utils.py: Utilities for data processing.
• technical_indicators.py: Adds technical indicators to the data.
• feature_engineering.py: Enhances data with technical indicators and normalizes it.
• logger.py: Sets up logging for the project.
• objective.py: Defines the objective function for hyperparameter optimization.
• model.py: Defines the LSTM neural network architecture.
• train.py: Trains and evaluates the model with historical data.
• train_model.py: Contains functions for model training.
• train_utils.py: Utility functions for training.
• predict.py: Uses the trained model for future price predictions.

Data

Data is downloaded from Yahoo Finance using yfinance. The raw data is stored in data/raw_data.csv.

Processed data files:

• data/processed_data_{TICKER}.csv: Cleaned up data.
• data/raw_data_{TICKER}.csv: Raw data.
• data/scaled_data_{TICKER}.csv: Normalized data with technical indicators for model training.

Usage

Training

Run the training script to train the model:

sh train.sh

Run training scripts individually

# Prepare the data
python3 src/data_preparation.py

# Setup the feature engineering
python3 src/feature_engineering.py

# Train the model
python3 src/train.py

Predict

Run the prediction script to make predictions:

sh predict.sh

# Predict the test data
python3 src/predict.py

Logging

The project uses Python's logging module to output logs to both the terminal and log files, aiding in monitoring and debugging.

Requirements

pandas~=2.1.3
numpy~=1.26.2
tensorflow~=2.16.1
featuretools~=1.31.0
statsmodels~=0.14.2
matplotlib~=3.9.0
scikit-learn~=1.3.2
keras~=3.3.3
joblib~=1.3.2
yfinance~=0.2.32
PyQt5~=5.15.10
optuna~=3.6.1
optuna-integration~=3.6.0

Install Dependencies

pip install -r requirements.txt

Contributors

• Gianluca Mazza
• Matteo Garbelli

License

This project is licensed under the MIT License - see the LICENSE file for details.

btc-prediction-lstm's Issues

hello ，how to get data to train model?