DeepStock📊

DeepStock is a deep learning framework for stock price prediction and forecasting using LSTMs and feature engineering.
It supports training, evaluation, and visualization of model performance with residuals, loss curves, and volatility indicators.

Features

• Preprocessing with MinMaxScaler and rolling features (MA10, MA20, MA50, returns).

• Sequence generation for supervised learning on time series.

• LSTM-based forecasting with:

  • Model checkpointing
  • Early stopping & learning rate scheduling

• Evaluation metrics:

  • MAE, RMSE, R², MAPE
  • Directional Accuracy (%)
  • Bias correction

• Visualizations:

  • Actual vs. Predicted prices with forecasts
  • Residuals (Actual − Predicted)
  • Smoothed training/validation loss curves
  • Volatility indicators (rolling std & Bollinger Band width)

---

📂 Project Structure

deepstock/
│
├── data/                  # input stock data (CSV or yfinance)
│
├── checkpoints/           # saved models (.keras)
│
├── results/               # predictions, plots, forecasts
│   ├── predictions.csv
│   ├── future_predictions_60d.csv
│   ├── future_predictions_*_6m.csv
│   └── plots/
│
├── images/                # example plots & animations
│   ├── example1.png
│   └── example2.gif
│
├── prediction.py          # main training + forecasting loop
├── requirements.txt       # dependencies
└── README.md              # project docs

---

Installation

Clone the repo:

git clone https://github.com/blkpvnthr/deepstock.git
cd deepstock

Create a virtual environment:

macOS/Linux

python -m venv .venv
source .venv/bin/activate

Windows

.venv\Scripts\activate

Install dependencies:

pip install -r requirements.txt

---

Usage

1. Run training + forecasting

   python prediction.py

This will:

• Train or load an existing model per ticker.
• Save checkpoints to checkpoints/.
• Generate predictions and forecasts (60-day & 6-month).
• Save plots into results/plots/.

---

Example Output

Actual + Predictions + Forecasts Residuals & Loss Curves Volatility Indicators

📊 Example Metrics

CGNX Evaluation Metrics:
MAE   : 0.0621
RMSE  : 0.0717
R²    : -1.1457
MAPE  : 48.67%
Directional Accuracy: 46.81%
Bias Correction: +1.235

Actual + Predictions + Forecasts

---

🛠 Requirements

Python 3.10+ pandas numpy scikit-learn matplotlib tensorflow / keras

Install them with:

pip install -r requirements.txt

---

🔮 Roadmap

Add hyperparameter tuning
Integrate additional technical indicators
Support transformer-based time-series models

---

📜 License

MIT License. Feel free to fork and adapt for your own trading experiments 🚀

---

✨ Acknowledgements

Inspired by real-world quantitative finance research and experimentation with deep learning on market data.