Stock Price Prediction using RNN-LSTM

A deep learning model that predicts Google stock prices using a Recurrent Neural Network (LSTM) architecture.

Overview

This project implements a stock price prediction system using a stacked LSTM (Long Short-Term Memory) neural network. The model uses historical stock prices to predict future price movements of Google stock.

Requirements

numpy
pandas
matplotlib
scikit-learn
keras
tensorflow

Project Structure

The project consists of three main parts:

Data Preprocessing

• Imports Google stock price training data
• Applies MinMaxScaler for feature scaling (0-1 range)
• Creates a 60-day timestep structure for predictions
• Reshapes data into 3D format required for LSTM

Model Architecture

The RNN model includes:

• 4 LSTM layers (50 units each)
• Dropout layers (20% dropout rate)
• Dense output layer
• Adam optimizer
• Mean squared error loss function

Prediction and Visualization

• Processes test data using the same scaling
• Makes predictions on test set
• Visualizes results comparing actual vs predicted prices

Key Features

Data Structure

• Uses 60 previous days for prediction
• 20 financial days per month
• 3-month prediction window

Best Practices

• Maintains consistent scaling between training and test data
• Uses proper data shapes for LSTM input
• Implements dropout for regularization

Usage

1. Prepare your data:

   • Training data: CSV file with stock prices
   • Test data: CSV file with validation data

2. Run the model:

# Train the model
regressor.fit(X_train, y_train, epochs=100, batch_size=32)

# Make predictions
predicted_stock_price = regressor.predict(X_test)

3. Visualize results:

plt.plot(real_stock_price, color='red', label='Real Google Stock Price')
plt.plot(predicted_stock_price, color='blue', label='Predicted Google Stock Price')
plt.show()

Model Parameters

• LSTM Units: 50 per layer
• Dropout Rate: 0.2
• Batch Size: 32s
• Epochs: 100

Notes

• The model uses only the 'Open' price as a predictor
• Additional features (volume, close price, etc.) can be added for potentially better predictions
• The prediction window is optimized for 60 days of historical data