Stock Price Prediction using ARIMA and Gradient Boosting For Invsto

Project Overview

This notebook implements a data science project to predict stock prices using machine learning models. The project aims to compare the performance of two different approaches:

1. ARIMA (Autoregressive Integrated Moving Average) Model: A traditional time series forecasting method.
2. Gradient Boosting Model: A powerful machine learning ensemble method that can capture complex patterns and leverage engineered features.

Dataset

• Source: Yahoo Finance, accessed using the yfinance Python library.
• Tickers: Initially downloaded data for AAPL, MSFT, GOOG, AMZN, TSLA, but the primary analysis focuses on AAPL.
• Data Type: Daily OHLC (Open, High, Low, Close) and Volume data.
• Time Period: January 1, 2018, to December 31, 2023.

Methodology

The notebook follows a standard data science pipeline:

1. Data Preparation:

   • Downloads historical stock data using yfinance.
   • Handles missing values using forward fill (fillna(method='ffill')).
   • Selects 'Close' prices for AAPL and ensures the index is in datetime format.

2. Exploratory Data Analysis (EDA):

   • Visualizes AAPL closing prices and trading volume over time.
   • Calculates and plots 50-day and 200-day moving averages.
   • Computes and visualizes daily returns.
   • Generates ACF (Autocorrelation Function) and PACF (Partial Autocorrelation Function) plots to aid in ARIMA parameter selection.

3. Feature Engineering:

   • Creates lagged features for 'Close', 'Volume', and 'Daily_Return' (lags of 1, 3, and 7 days).
   • Calculates 7-day and 30-day rolling means of the 'Close' price (shifted to avoid look-ahead bias).
   • Computes the percentage change in volume.

4. Modeling:

   • ARIMA Model:
     • Performs stationarity checks using the Augmented Dickey-Fuller (ADF) test.
     • Applies differencing to make the time series stationary.
     • Uses ACF and PACF plots to guide the selection of ARIMA parameters (p, d, q).
     • Trains an ARIMA(1, 1, 1) model (example parameters).
     • Forecasts future prices and inverts differencing to obtain forecasts in the original scale.
   • Gradient Boosting Model:
     • Prepares data with engineered features for the Gradient Boosting model.
     • Splits data into training and testing sets.
     • Scales features using MinMaxScaler.
     • Performs hyperparameter tuning using GridSearchCV to find optimal parameters for GradientBoostingRegressor.
     • Trains the best Gradient Boosting model.
     • Predicts stock prices on the test set.

5. Model Evaluation:

   • Evaluates both ARIMA and Gradient Boosting models using the following metrics:
     • RMSE (Root Mean Squared Error)
     • MAE (Mean Absolute Error)
     • MAPE (Mean Absolute Percentage Error)
   • Compares the performance of the two models based on these metrics.

6. Report and Presentation (Conceptual - Notebook includes code for analysis and results):

   • The notebook is structured to facilitate the creation of a comprehensive report and presentation (outline provided in the notebook comments and previous responses).

Results

The model evaluation yielded the following results for predicting AAPL stock prices on the test set:

Model Evaluation Metrics:

Metric	ARIMA Model	Gradient Boosting Model
RMSE	19.76	8.93
MAE	17.23	6.17
MAPE	9.83%	3.40%

Key Findings:

• Gradient Boosting significantly outperformed the ARIMA model across all evaluation metrics (RMSE, MAE, MAPE).
• The Gradient Boosting model achieved a much lower MAPE of 3.40% compared to ARIMA's 9.83%, indicating a considerably better relative accuracy in price prediction.
• These results suggest that for AAPL stock price prediction in this context, a non-linear model like Gradient Boosting, leveraging engineered features, is more effective than a traditional linear time series model like ARIMA.

Visualizations:

The notebook includes various visualizations to support the analysis, including:

• Time series plots of AAPL closing price and volume.
• Plots of closing price with moving averages.
• Daily returns plot.
• ACF and PACF plots for ARIMA analysis.
• Forecast plots for both ARIMA and Gradient Boosting models compared to actual prices.

Technologies Used

• Python Libraries:

  • yfinance: For downloading stock market data.
  • pandas: For data manipulation and analysis.
  • numpy: For numerical operations.
  • statsmodels: For ARIMA modeling and time series analysis.
  • scikit-learn: For Gradient Boosting, model evaluation, data splitting, and feature scaling.
  • matplotlib and seaborn: For data visualization.

• Environment: Google Colab (cloud-based Python environment).

Usage

To run this notebook in Google Colab:

1. Open in Colab: Click on "Open in Colab" button (if available) or upload the notebook file to your Google Drive and open it in Google Colab.
2. Install Libraries (if necessary): The notebook includes !pip install commands at the beginning to install required libraries if they are not already present in your Colab environment. Run these cells first if needed.
3. Execute Cells Sequentially: Run the notebook cells in order from top to bottom. You can use Shift+Enter to run a cell and move to the next one.
4. Observe Outputs and Visualizations: Check the printed outputs and generated plots after running each section of the code.