MarketMind: Stock Price Prediction with PyTorch/LSTMs

This project applies deep learning techniques (PyTorch) to forecast short-term stock index movements using NASDAQ from December 2024. The pipeline covers preprocessing, model training, evaluation, and visualization. Future work aims to incorporate a visual analytics dashboard and parse stock market data from the web using Requests and Pandas.

---

Features

• Preprocessing of historical stock data (CSV from Yahoo Finance)
• LSTM-based predictive model built with PyTorch
• Modularized code (src/ folder with training, evaluation, prediction, and utilities)
• Training and evaluation logs with clear metrics
• Visualizations:
  • Training vs Validation Loss Curves
  • Actual vs Predicted Closing Prices
• Reproducible results with requirements.txt and config.yaml

---

Results

Training vs Validation Loss

Actual vs Predicted Closing Prices

These results show that the model converged smoothly (validation loss ~0.0125 by epoch 10) and achieved a mean absolute error (MAE) under ~80 points on the NASDAQ index, which is strong for such a short dataset.

---

Future Work

• Extend dataset beyond Dec 2024 for better generalization
• Add hyperparameter tuning with Optuna
• Build a Streamlit dashboard for interactive forecasting