📈 LSTM Stock Price Prediction

A comprehensive deep learning framework for stock price prediction using LSTM neural networks with advanced feature engineering, automated hyperparameter optimization, and interactive trading simulation.

✨ Features

• LSTM-Based Time Series Forecasting: LSTM specifically designed for sequential financial data
• Automated Hyperparameter Optimization: Grid search across multiple hyperparameters with early stopping
• Automated Data Retrieval Pipeline: yfinance and fredapi integration to retrieve up-to-date market and economic data
• Feature Engineering Pipeline: Technical indicators (MACD, RSI, ATR) and macroeconomic data integration
• Interactive Dashboard: Real-time experiment configuration with Feature Selection and Hyperparameter Search Configuration in a visual Dashboard UI
• Trading Strategy Simulation: Backtest several different trading strategies with detailed performance metrics
• Comprehensive Visualization: Training progress, predictions, error analysis, and equity curves
• Flexible Model Persistence: Save and load trained models with configurations

🏗️ Architecture

🚀 Getting Started

Prerequisites

• Python 3.8+
• CUDA-compatible GPU (optional, but recommended)
• PyTorch with CUDA support

Installation

1. Clone the repository:

git clone <repository-url>
cd LSTMPrediction

2. Create a virtual environment:

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

Data Setup

Place your financial data in the data/ directory. Expected format:

• CSV files with date index
• Columns: Close_{TICKER}, Volume_{TICKER}, technical indicators, macroeconomic data

OR retrieve data using the provided pipeline:

data_acquision.ipynb

📊 Usage

1. Quick Start with Jupyter Notebook

Open StockPricePrediction.ipynb and run through the cells:

2. Interactive Dashboard

Launch the interactive dashboard for experiment configuration:

Configure:

• Training Features: Select from technical indicators and macro features
• Hyperparameter Grid: Window size, batch size, learning rate, dropout, etc.
• Model Architecture: Hidden layers, LSTM layers, optimizer selection

3. Feature Selection

The framework supports incremental feature selection:

This tests:

1. Target feature only (baseline)
2. Target + each additional feature individually
3. All features combined

4. Model Evaluation

Generates:

• Training history plots
• Predictions vs actuals on test set
• Residual analysis
• Performance metrics (MSE, MAE, R²)

5. Trading Strategy Backtesting

🔧 Configuration

📈 Supported Features

Technical Indicators

• Price data: Close, Volume
• Momentum: MACD, RSI
• Volatility: ATR, VIX

Macroeconomic Data

• MACRO_cpi: Consumer Price Index
• MACRO_rate: Interest rates
• MACRO_ppi: Producer Price Index
• MACRO_gdp: GDP
• MACRO_unrate: Unemployment rate
• MACRO_CSI: Consumer Sentiment Index
• MACRO_HS: Housing Starts

🎯 Model Training Pipeline

1. Data Preparation: Load, clean, and scale data
2. Feature Engineering: Calculate technical indicators
3. Sequence Creation: Generate sliding windows for LSTM input
4. Train/Val/Test Split: 80%/10%/10% split
5. Grid Search: Automated hyperparameter optimization
6. Early Stopping: Prevent overfitting with validation monitoring
7. Model Selection: Choose best model based on validation loss
8. Evaluation: Test set performance and visualization
9. Trading Simulation: Backtest with realistic trading scenarios

📁 Project Structure

LSTMPrediction/
├── classes/              # Core framework classes
├── data/                 # Financial datasets
├── run/                  # Experiment outputs and checkpoints
├── StockPricePrediction.ipynb  # Main training notebook
├── data_acquisition.ipynb      # Data collection pipeline
├── simulation.ipynb            # Trading simulations
└── requirements.txt            # Python dependencies

🔬 Experiment Tracking

All experiments are automatically saved to run/ with:

• Model checkpoints (checkpoint.pt)
• Configuration files (config.json)
• Training plots
• Prediction visualizations
• Performance metrics CSV

🎨 Visualization

The framework provides comprehensive visualizations:

• Training Progress: Real-time loss curves with early stopping indicators
• Predictions: Overlaid predicted vs actual values on test set
• Residual Analysis: Error distribution and patterns
• Equity Curves: Trading simulation performance over time
• Feature Analysis: SHAP values for feature importance