Crypto Currency Price Prediction

Desktop application for cryptocurrency price forecasting with recurrent neural networks (LSTM, GRU, and LSTM+GRU). This project is educational and should not be treated as trading advice.

Overview

The project uses a Tkinter GUI (MVC structure) to: - select a market (BTC, ETH, DOGE, LTC), - choose a model architecture, - train on online or local CSV data, - visualize predicted vs. actual prices, - calculate simple gain over a selected window.

Training runs in a background thread. Progress and ETA are shown in: - terminal logs, - GUI status line.

Tech Stack

• Python >=3.10,<3.13
• TensorFlow / Keras
• NumPy, Pandas, scikit-learn
• Matplotlib
• yfinance + pandas-datareader
• Tkinter (Azure theme)
• uv (dependency and environment management)

Project Structure

.
|-- app.py                 # Application entry point (Controller)
|-- app/
|   |-- model.py           # Data access, training, prediction, gain
|   |-- view.py            # Tkinter GUI
|   |-- azure.tcl          # GUI theme
|   |-- test.csv           # Default local CSV for offline mode
|   |-- assets/
|   `-- images/
|-- DATA/
|   `-- download.py        # Utility script for fetching CSV data
|-- DOC/                   # Report and presentation materials
|-- pyproject.toml
`-- uv.lock

Quick Start

1. Install dependencies and run the app:

uv sync
uv run python app.py

Using the App

1. Choose currency, data source, and model type.
2. Configure Prediction days, Future days, and Plot range.
3. Click Train.
4. Monitor progress in terminal and GUI status line.
5. Click Plot to show the chart.
6. Click GAIN to compute percent change.

Data and Date Semantics

• Prediction days: lookback window size (how many past observations are used as input).
• Future days: forecast horizon in days (for example, 1 means predict one day ahead).
• Plot range: number of dated points displayed on the chart.
• Chart X-axis now uses concrete calendar dates (YYYY-MM-DD).
• Predicted series is shifted by Future days, so each predicted point is plotted at its target date.

Showcase

GUI

The application features a clean desktop interface built with Tkinter (Azure theme), organized into four panels:

• Select Currency — radio buttons to choose between BTC, ETH, Doge, and LTC.
• Online Source — selects the data provider: Yahoo, Stooq, or Naver.
• Select Model — picks the neural network architecture: LSTM, GRU, or LSTM+GRU.
• Settings panel (top-right) — sliders for Prediction lookback and data window (Based on last 30 days), a toggle to enable/disable the online database, and spinboxes for Future days and Plot range.
• Action buttons — Train starts model training in a background thread, Plot renders the forecast chart, GAIN computes percent price change, Select file loads a local CSV for offline mode, and HELP opens documentation.

Prediction Chart

After training, clicking Plot opens a Matplotlib chart comparing Actual Prices (cyan) against Predicted Prices (dark purple) over the selected date range. The predicted series is shifted forward by Future days, so each predicted point aligns with its target date. The example below shows a BTC forecast over a 20-day window, where the model closely tracks the sharp price rise from ~84 000 to ~94 000 USD.

License

This project is licensed under the MIT License. See LICENSE.