🛒 Walmart Sales Analysis & Forecast Dashboard

🖼️ Final StreamLit Result

A full machine learning pipeline and interactive dashboard to analyze and forecast Walmart sales. Combines model experimentation with an intuitive Streamlit app for exploring weekly sales trends and predictions.

---

📋 Table of Contents

1. 📌 Project Overview
2. 📊 Dataset
3. 🔄 Pipeline & Workflow
4. 📈 Insights
5. 💻 Usage Examples
6. 🤝 Contributing
7. 📬 Contact

---

📌 Project Overview

This project analyzes Walmart’s sales data and forecasts future weekly sales using machine learning. It includes:

• A Jupyter Notebook that trains multiple models (Random Forest, XGBoost, LightGBM, ...) using a custom time-series pipeline.
• A Streamlit dashboard to explore sales data and compare predictions interactively, allowing users to filter by store, department, time range and also ranking from best to worst based on total sales, $ Growth and % Growth.

---

📊 Dataset

Sources:

• Model Training: Historical sales and market data from https://www.kaggle.com/competitions/walmart-recruiting-store-sales-forecasting/overview, yfinance, pandas-datareader, and akshare. Also some columns for holidays and special events or Tax return used a ramp up and down do represent importance.
• Dashboard: Pre-processed CSV file df_wm_store_sales_predictions.csv, which contains weekly sales and predictions.

Data Type	Source	Description
Stock Data	yfinance / pandas-datareader	Weekly Walmart stock prices and economic indicators
Processed Dataset	Local CSV (df_wm_store_sales_predictions.csv)	Sales & predictions used in the dashboard

---

🔄 Pipeline & Workflow

🧠 Model Training (Jupyter Notebook)

1. Data Fetching – Collect WMT stock and macroeconomic indicators
2. Feature Engineering – Create time-aware features and lags
3. Time-Series CV – TimeSeriesSplit with performance tracking
4. Modeling – Train Random Forest, XGBoost, LightGBM, ...
5. Interpretation – SHAP values and permutation importance ( not completed because of lack of compute power and time constrains

📊 Streamlit Dashboard

• app.py: Main controller for layout, interaction, routing
• data_loader.py: Loads cached data using @st.cache_data
• filters.py: Applies store/department/date filters
• metrics.py: Calculates KPIs (sales totals, growth, date ranges)
• ui_components.py: Charts, grids, headers, KPIs, footers

---

📈 Insights

• The Random Forest model accurately captures general sales trends.
• SHAP analysis would highlights the most impactful features on predictions.
• Dynamic visual tools make it easy to identify underperforming stores or departments.

Next Steps

• 🔧 Hyperparameter Tuning: Optimize Random Forest for better accuracy ( if possible get more data on departments and have daily or hourly sales instead of weekly)
• 📊 More Visuals: Add SHAP force plots

---

💻 Usage Examples

Install requirements:

• For notebook usage just ran all cells. If you want to run on kaggle just delete the """ in the first cell. ( note that there is a function called "%%skip". This function was used to run all cell but those that start with that.
• To check streamlit app just run the streamlit.bat found in the main directory.

---

🤝 Contributing

1. Fork the repo
2. Create a feature branch (git checkout -b feature/YourFeature)
3. Commit your changes (git commit -m "Add new analysis")
4. Push (git push origin feature/YourFeature)
5. Open a Pull Request

---

📬 Contact

Authors

Jorge M. M. L. Rodrigues

• Email: [email protected]
• GitHub: github.com/JorgeMMLRodrigues

Feel free to open issues or discussions!