LSTM-Based Stock Price Prediction Using Deep Learning

This repository contains a detailed implementation of a stock price prediction model leveraging Long Short-Term Memory (LSTM) networks. The project is designed to predict future stock prices based on historical data and provides visualizations to analyze the model's performance.

---

Introduction

Stock price prediction is a challenging problem due to the volatility and non-linear trends in financial markets. This project implements an LSTM-based neural network, which is well-suited for time-series analysis. The model is trained on 20 years of historical stock data from Google, obtained using the Yahoo Finance API, to predict future closing prices.

---

Features

1. Data Collection:

   • Utilizes Yahoo Finance API to fetch historical stock data.
   • Processes stock data for analysis and model training.

2. Data Analysis:

   • Performs exploratory data analysis (EDA) with descriptive statistics and visualizations.
   • Plots key metrics like adjusted close price, moving averages, and percentage changes.

3. Model Development:

   • Constructs an LSTM-based model using TensorFlow/Keras.
   • Implements data normalization and sliding window mechanisms for time-series data preparation.

4. Visualization:

   • Provides comparative plots for original vs. predicted prices.
   • Includes visualizations of stock trends and moving averages.

5. Performance Evaluation:

   • Calculates the Root Mean Square Error (RMSE) to evaluate model accuracy.
   • Offers insights into model predictions and real stock price trends.

---

Dataset

The dataset comprises 20 years of historical stock data for Google (GOOG) obtained from the Yahoo Finance API. The key features include:

• Open: Opening price of the stock.
• High: Highest price of the stock during the trading session.
• Low: Lowest price of the stock during the trading session.
• Close: Closing price of the stock.
• Adj Close: Adjusted closing price accounting for splits and dividends.
• Volume: Number of shares traded.

---

Model Architecture

The LSTM model consists of:

• Input layer for time-series data (100 timesteps).
• Two LSTM layers:
  • First LSTM layer with 128 units and return sequences enabled.
  • Second LSTM layer with 64 units and no return sequences.
• Dense layers for output:
  • One Dense layer with 25 units.
  • Final Dense layer with 1 unit to predict the next stock price.

Optimizer: Adam
Loss Function: Mean Squared Error (MSE)

---

Results

• The model achieved an RMSE of 3.76 on the test dataset.
• Visualizations highlight the alignment between predicted and actual stock prices, showcasing the effectiveness of the LSTM model.

---