Store Sales Forecasting using Prophet

This project focuses on forecasting store sales using Facebook Prophet, a powerful time series forecasting tool developed by Meta. The dataset used in this project is from Kaggle, where the original analysis was done using ARIMA. In this project, we use Prophet to predict future sales based on historical data.

---

Table of Contents

1. Project Overview
2. Dataset
3. Project Structure
4. Installation
5. Usage
6. Configuration
7. Results
8. Contributing
9. License

---

Project Overview

The goal of this project is to predict future store sales using historical sales data. The project includes the following steps:

1. Data Preprocessing: Load and clean the dataset, handle missing values, and prepare the data for time series analysis.
2. Exploratory Data Analysis (EDA): Visualize sales, quantity, and profit distributions, and analyze monthly sales trends.
3. Seasonal Decomposition: Decompose the time series into trend, seasonality, and residuals to understand underlying patterns.
4. Sales Forecasting: Use Facebook Prophet to forecast future sales and visualize the results.
5. Model Evaluation: Evaluate the model's performance using metrics like MAE (Mean Absolute Error) and RMSE (Root Mean Squared Error).

---

Dataset

The dataset used in this project is from Kaggle. It contains historical sales data for a retail store, including the following columns:

• Order Date: The date of the order.
• Sales: The total sales for the order.
• Quantity: The quantity of items sold.
• Profit: The profit generated from the order.

The dataset is stored in the data/ directory as stores_sales_forecasting.csv.

---

Project Structure

The project is structured as follows:

Prophet/
├── data/
│   └── stores_sales_forecasting.csv       # Dataset
├── notebooks/
│   └── analysis.ipynb                     # Jupyter notebook for analysis
├── src/
│   ├── data_preprocessing.py              # Data loading and preprocessing
│   ├── data_visualization.py              # Visualization functions
│   ├── prophet_forecasting.py             # Prophet forecasting implementation
│   ├── seasonal_decomposition.py          # Seasonal decomposition
│   └── utils.py                           # Utility functions
├── config.py                              # Configuration for Prophet
├── requirements.txt                       # Python dependencies
└── README.md                              # Project documentation

---

Installation

To set up the project, follow these steps:

1. Clone the repository:

   git clone https://github.com/your-username/Prophet-Sales-Forecasting.git
   cd Prophet-Sales-Forecasting

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Jupyter notebook:

   jupyter notebook notebooks/analysis.ipynb

---

Usage

1. Data Preprocessing

The data_preprocessing.py script loads and preprocesses the dataset. It performs the following tasks:

• Drops columns with unique or single values.
• Converts the Order Date column to a datetime format.
• Aggregates sales data into monthly sales.

2. Data Visualization

The data_visualization.py script provides functions to visualize:

• Sales, quantity, and profit distributions.
• Monthly sales trends.

3. Seasonal Decomposition

The seasonal_decomposition.py script decomposes the time series into trend, seasonality, and residuals using statsmodels.

4. Prophet Forecasting

The prophet_forecasting.py script implements the Prophet model for sales forecasting. It includes:

• Model initialization and training.
• Future sales prediction.
• Visualization of historical data and forecast.

5. Running the Analysis

The analysis.ipynb notebook combines all the steps:

• Load and preprocess the data.
• Visualize the data.
• Perform seasonal decomposition.
• Run Prophet forecasting and display results.

---

Configuration

The config.py file contains the configuration for the Prophet model. The default settings are:

PROPHET_CONFIG = {
    'yearly_seasonality': True,       # Enable yearly seasonality
    'weekly_seasonality': False,      # Disable weekly seasonality
    'daily_seasonality': False,       # Disable daily seasonality
    'seasonality_mode': 'multiplicative',  # Seasonality mode
    'forecast_periods': 24            # Number of months to forecast
}

You can modify these settings to adjust the model's behavior.

---

Results

1. Sales, Quantity, and Profit Distributions

The following visualizations show the distributions of Sales, Quantity, and Profit from the dataset:

• Sales Distribution: The sales data is right-skewed, indicating that most sales are concentrated in the lower range, with a few high-value outliers.
• Quantity Distribution: The quantity of items sold follows a similar pattern, with most orders involving a small number of items.
• Profit Distribution: The profit distribution shows both positive and negative values, with most transactions yielding small profits or losses.

---

2. Monthly Sales Trend

The monthly sales trend shows the total sales over time, aggregated by month:

• The plot reveals a clear upward trend in sales from 2014 to 2017, with some seasonal fluctuations.
• Peaks in sales are observed around mid-year, suggesting potential seasonality in the data.

---

3. Seasonal Decomposition

The time series is decomposed into Trend, Seasonality, and Residuals to better understand the underlying patterns:

• Trend: The trend component shows a steady increase in sales over time, with a slight dip in 2016.
• Seasonality: The seasonal component reveals a repeating pattern, with peaks in mid-year and troughs at the beginning and end of the year.
• Residuals: The residuals represent the noise in the data after removing the trend and seasonality. They are relatively small, indicating that the model captures most of the variability.

---

4. Sales Forecast with Prophet

The Prophet model is used to forecast future sales. The forecast includes historical data, predicted values, and uncertainty intervals:

• Historical Data: The blue line represents the actual sales data from 2014 to 2017.
• Forecast: The green line shows the predicted sales for the next 24 months (2018–2020).
• Uncertainty Intervals: The shaded green area represents the confidence interval for the forecast, indicating the range within which the actual sales are likely to fall.

---

5. Trend and Seasonality Components

The following plots show the Trend and Yearly Seasonality components extracted by the Prophet model:

• Trend: The trend component confirms the steady growth in sales over time, with a slight dip in 2016.
• Yearly Seasonality: The seasonality plot shows a clear yearly pattern, with sales peaking in the middle of the year (around July) and dropping at the beginning and end of the year.

---

Contributing

Contributions are welcome! If you'd like to contribute, please follow these steps:

1. Fork the repository.
2. Create a new branch (git checkout -b feature/YourFeature).
3. Commit your changes (git commit -m 'Add some feature').
4. Push to the branch (git push origin feature/YourFeature).
5. Open a pull request.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Acknowledgments

• Dataset: Kaggle Store Sales Forecasting
• Facebook Prophet: Prophet Documentation