🧭 Sales Forecasting & Rolling Holt-Winters (Rossmann Case Study)

This project demonstrates time series forecasting using both rolling and direct prediction strategies based on the Holt-Winters Exponential Smoothing model. It applies to the Rossmann Store Sales dataset and includes data exploration, automatic trend & seasonality detection, forecast evaluation, and automated PDF reporting.

---

📊 Project Overview

The main objective of this project is to:

• Analyze sales patterns at both company and store level.
• Detect trend and seasonality automatically.
• Perform rolling forecasts (iterative 7-day predictions) and direct forecasts (30-day batch predictions).
• Evaluate performance using multiple statistical metrics.
• Generate a PDF report summarizing results and metrics.

---

🧠 Key Features

• 📈 Exploratory Data Analysis (EDA):
  Includes visualizations of trends, seasonality, promotions, store types, and competition distance.

• 🔍 Time Series Analysis:
  Automatically detects trend and seasonality using ACF and decomposition.

• 🔄 Forecasting Strategies:

  • Rolling Forecast: Iterative 7-day forecast updated every cycle.
  • Direct Forecast: One-time 30-day prediction.
    Both trained using Holt-Winters Exponential Smoothing.

• 🧮 Performance Evaluation:
  Results include both absolute and relative error metrics:

  Metric	Description	Unit
  RMSE	Root Mean Squared Error	Absolute
  Relative RMSE	RMSE as % of mean actuals	%
  MAE	Mean Absolute Error	Absolute
  Relative MAE	MAE as % of mean actuals	%
  MAPE	Mean Absolute Percentage Error (Company-level only)	%
  R²	Coefficient of Determination	-

• 📄 Automated Reporting:
  Creates clean, professional reports for both company and store forecasts, including daily forecasted values and performance summary.

• 🏬 Store-Level Analysis:
  Includes example forecasting for an individual store (Store ID = 1).

---

🧩 Project Structure

📁 data/
 ├── rossmann/
 │   ├── train.csv
 │   ├── store.csv
📁 report/
 ├── sales_forecast_report_company.pdf
 ├── sales_forecast_report_store1.pdf
 ├── forecast_comparison_company.png
📄 Sales_Forecasting.ipynb
📄 requirements.txt
📄 README.md

---

⚙️ Installation & Setup

1. Clone this repository

   git clone https://github.com/ghfri-code/Sales-Forecasting.git
   cd Sales-Forecasting

2. Install dependencies

   pip install -r requirements.txt

3. Add dataset
   Download the Rossmann Store Sales dataset and place it under:

   data/rossmann/train.csv
   data/rossmann/store.csv
   

4. Run the notebook

   jupyter notebook Sales_Forecasting.ipynb

---

📈 Example Visualizations

• Daily sales trend with a 30-day moving average
• Sales by day of week and monthly sales trend
• Top 10 stores by sales
• Promotion impact analysis
• Store type & competition distance analysis

---

🧮 Forecasting Process

1. Data Preparation: Load and merge datasets; handle missing days and promotions
2. Trend & Seasonality Detection: Auto-detect seasonal period using ACF and seasonal_decompose
3. Forecast Generation: Apply both rolling and direct Holt-Winters forecasting (statsmodels.tsa.holtwinters.ExponentialSmoothing)
4. Model Evaluation: Compute RMSE, MAE, MAPE, and R² for comparison
5. Visualization & Reporting: Save forecast plot as PNG and generate PDF summary report

---

🧰 Tech Stack

• Python 🐍
• pandas, numpy, matplotlib, seaborn – data handling & visualization
• statsmodels – time series modeling (Holt-Winters)
• scikit-learn – evaluation metrics
• fpdf – PDF report generation

---

📜 Requirements

pandas
numpy
matplotlib
statsmodels
scikit-learn
fpdf

---

🚀 Future Improvements

• Integrate Prophet and XGBoost for hybrid forecasts
• Add automated hyperparameter tuning for seasonal components
• Implement automated parameter tuning for seasonal periods
• Add interactive dashboards (Streamlit / Plotly)