Predicting weekly department-level sales across 45 Walmart stores using historical markdown and holiday data — deployed as a live Flask web application.

Retail chains like Walmart face a critical challenge: accurately forecasting sales to optimize inventory, staffing, and supply chain decisions — especially during high-impact holiday seasons.

This project builds an end-to-end machine learning solution to predict weekly store sales using 2+ years of historical data from 45 Walmart stores across different US regions. The model accounts for promotional markdowns, seasonal spikes (Super Bowl, Thanksgiving, Christmas, Labour Day), and store-level features.

• Helps Walmart plan stock levels and warehouse space before high-demand periods
• Identifies which stores and departments are most affected by holiday markdowns
• Enables data-driven decisions on regional promotions and resource allocation

Source: Kaggle — Walmart Store Sales Forecasting

Key Challenge: Holiday weeks are weighted 5x higher in evaluation (WMAE metric), making accurate holiday prediction critical.

Walmart-Store-Sales-Forecasting/
├── data/
│   ├── train.csv
│   ├── test.csv
│   ├── features.csv
│   └── stores.csv
├── templates/
│   └── index.html          # Flask frontend
├── static/
│   └── style.css
├── Walmart Store Sales Forecasting.ipynb   # Full EDA + Modelling
├── app.py                  # Flask application
├── model.pkl               # Serialized trained model
├── requirements.txt
└── Dockerfile

• Analyzed seasonal patterns in weekly sales across stores and departments
• Identified holiday spikes — Thanksgiving consistently outperforms Christmas in actual sales
• Found that January sales drop sharply post-holiday season
• Examined impact of external factors: CPI, fuel price, unemployment, temperature

• Extracted time features: Week, Month, Year from date
• Engineered holiday flags for Super Bowl, Labour Day, Thanksgiving, Christmas
• Merged store metadata (type, size) with weekly sales and external features
• Handled missing MarkDown values (up to 4,140 nulls in some columns) using median imputation

• Evaluated multiple regression approaches for time-series forecasting
• Applied Random Forest Regressor as the primary model due to its ability to capture non-linear relationships and feature interactions
• Used cross-validation with time-aware splits to prevent data leakage

WMAE (Weighted Mean Absolute Error) — holiday weeks weighted 5x:

WMAE = (1 / Σwᵢ) × Σ wᵢ |yᵢ - ŷᵢ|

• Serialized model using pickle → model.pkl
• Built a Flask web app for real-time sales prediction
• Containerized with Docker for portability
• Deployed on web (previously Heroku, migrating to Render)

Key Finding: Holiday markdown events — especially Thanksgiving — are the single strongest predictor of sales spikes. Store type and size also significantly influence baseline weekly sales.

# Clone the repository
git clone https://github.com/vicky60629/Walmart-Store-Sales-Forecasting.git
cd Walmart-Store-Sales-Forecasting

# Install dependencies
pip install -r requirements.txt

# Run the Flask app
python app.py

Then open http://localhost:5000 in your browser.

docker build -t walmart-forecasting .
docker run -p 5000:5000 walmart-forecasting

Input Form	Prediction Output

What worked well:

• Feature engineering around holidays significantly improved prediction accuracy
• Random Forest handled the non-linear interactions between store size, type, and markdown events effectively

Future enhancements:

• Integrate XGBoost / LightGBM for potential performance gains
• Add SARIMA / Prophet for pure time-series baseline comparison
• Build interactive dashboard with Streamlit or Plotly Dash
• Incorporate external data: weather APIs, economic indicators
• Retrain with MLflow for experiment tracking

Vicky Gupta — Data Engineering Analyst @ Accenture (4.5 years) | Aspiring Data Scientist

Skilled in PySpark, ETL pipelines, and end-to-end ML systems. Passionate about building data products that solve real business problems.

🔗 LinkedIn | GitHub

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

⭐ If you found this project useful, please consider starring the repository!