📈 LSTM-Based Stock Forecasting for the Philippines' Top 16 Companies

🏆 Project Overview

This project utilizes a Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) to forecast stock price movements of the largest companies in the Philippines. The model predicts the next 7 days of stock prices based on over 10 years of historical data per company.

📅 Timeline

• March 2025

🛠 Features

✅ Python-based ETL Pipeline: Processed 100,000+ data points for efficient analysis and storage.
✅ 16 LSTM Models: Each trained on an individual company’s historical data.
✅ Time-Series Forecasting: Predicts the next 7 business days' stock prices (March 17 to March 25, 2025).
✅ Performance Evaluation: Mean error calculations for model assessment.

📊 Model Performance Summary

Metric	Value
Average MAE	18.49
Average MSE	1232.56
Average RMSE	20.99
Average MAPE	5.64%

📌 Interpretation: With an average MAPE of 5.64%, the model provides a fairly accurate prediction, with errors averaging around 5.64% of actual stock prices.

📂 Project Structure

📁 ForecastNotebooks/  # Jupyter notebooks for LSTM implementation
📁 StockData/          # Raw stock data since 2010 (if available), some from 2011-2012
📁 StockDataClean/     # Normalized stock data for model training
📁 StockMeanError/     # Contains TXT files with MAE, MSE, RMSE, and MAPE results for each company
📁 StockMerge/         # Merged datasets of predictions and cleaned stock data
📁 StockModels/        # Trained LSTM models (.h5 files)
📁 StockPredictions/   # 7-day stock price predictions (March 17 - March 25, 2025)
📁 StockTruePrices/    # Actual stock prices for comparison (March 17 - March 25, 2025)
📄 main.py            # LSTM training and forecasting script
📄 evaluate.py        # Computes mean error metrics
📄 README.md          # Project documentation

🚀 How to Run

1. Clone the Repository:

   git clone https://github.com/your-username/lstm-stock-forecasting.git
   cd lstm-stock-forecasting

2. Install Dependencies:

   pip install -r requirements.txt

3. Run Forecasting Model:

   python main.py

4. Evaluate Performance:

   python evaluate.py

📌 Future Improvements

🔹 Enhance feature engineering with technical indicators (e.g., RSI, MACD, Bollinger Bands).
🔹 Optimize hyperparameters to further reduce prediction errors.
🔹 Experiment with hybrid models (CNN-LSTM, Transformer-based forecasting).