Enhanced LSTM-ELU for Stock Market Forecasting

This project implements an enhanced LSTM model with ELU activation for stock market forecasting, as described in the research paper. The full-stack application includes a Python backend for model training/prediction and a Next.js frontend for visualization.

Features

• LSTM neural network with ELU activation function
• Stock data collection from Yahoo Finance
• Model training and evaluation pipeline
• Interactive visualization of predictions
• Comparison between baseline LSTM and enhanced LSTM-ELU

Tech Stack

Backend:

• Python 3.13+
• PyTorch (Deep Learning)
• yfinance (Data Collection)
• pandas (Data Processing)
• numpy (Numerical Operations)
• matplotlib (Visualization)
• FastAPI (API)

Frontend:

• Next.js 15
• React
• Apexchart/React-chart(Data Visualization)

Installation

1. Clone the repository:

git clone https://github.com/whoisbel/Exponential-Long-Short-Term-Memory
cd Exponential-Long-Short-Term-Memory

Install UV Project and Package Manager

Open your terminal and run the appropriate command based on your operating system:

Windows (PowerShell):

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Linux/macOS (Shell):

curl -LsSf https://astral.sh/uv/install.sh | sh

⚠️ Make sure you have curl installed on Linux/macOS. On Windows, run PowerShell as Administrator if needed.

3. Set up the backend:

cd backend
uv sync

4. Set up the frontend:

cd ../frontend
npm install

Configuration

Modify the model parameters in backend/src/scripts/config.json:

{
  "60-20_split_baseline": {
    "train_size": 0.6,
    "test_size": 0.2,
    "validation_size": 0.2,
    "hidden_sizes": [124],
    "dropout": 0.3,
    "seq_len": 28,
    "batch_size": 64,
    "epochs": 25,
    "learning_rate": 0.0015,
    "patience": 15
  },
  "80-20_split_enhanced": {
    "train_size": 0.8,
    "test_size": 0.2,
    "hidden_sizes": [50, 50, 50],
    "dropout": 0.2,
    "seq_len": 60,
    "batch_size": 128,
    "epochs": 100,
    "learning_rate": 0.001,
    "patience": 15
  }
}

Usage

1. Run the backend server:

cd backend
uv run uvicorn main:app --reload

2. Run the frontend development server:

cd ../frontend
npm install
npm run dev

3. Access the application at http://localhost:3000

Data Collection

The application automatically fetches stock data from Yahoo Finance for:

• Air Liquide (AI.PA)
• Date range: January 2010 - December 2024
• OHLCV (Open, High, Low, Close, Volume) data

Model Training

The system provides two models:

1. Baseline LSTM: Traditional LSTM with Tanh activation
2. Enhanced LSTM-ELU: Modified LSTM with ELU activation

Training metrics tracked:

• Mean Absolute Error (MAE)
• Root Mean Squared Error (RMSE)
• R² Score

Results

Based on the research paper, the enhanced LSTM-ELU model achieved:

• 75.64% reduction in MAE (1.90 vs 7.78)
• 75.37% reduction in RMSE (2.49 vs 10.10)
• R² score of 0.9852 (vs 0.7566 for baseline)

Acknowledgments

This implementation is based on the research paper: "Enhanced Long Short-Term Memory with Exponential Linear Unit for Stock Market Forecasting" by Dutaro, Belciña, and Cañedo (University of Mindanao, 2025)