Hamza Salah

Retail Demand Forecasting

Summary

Retail Demand Forecasting demonstrates how statistical modeling and time series analysis can transform retail decision-making. Incorporating seasonality and promotional data through SARIMAX enables accurate, actionable demand predictions-driving smarter inventory management and higher profitability.

Visualizations & Tables

Key Insights

SARIMAX models achieved the lowest RMSE (165.10), outperforming ARIMA and Auto ARIMA by accounting for seasonality and promotions.
Promotions had a measurable positive effect on sales volume, especially during peak seasons.
Toys showed steady growth, while Electronics and Apparel exhibited cyclical demand patterns.
Accurate forecasts can directly reduce overstocking, stockouts, and inventory carrying costs.

Code and Data

Dataset:
- Synthetic Retail Sales Dataset — Kaggle
- 3 years of daily sales data across multiple product categories
Data Cleaning & Preparation:
- Converted and standardized date fields
- Created time-based features (Year, Month, Year-Month)
- Aggregated daily sales to monthly totals
- Split dataset into training (pre-2024) and testing (2024) sets
View Jupyter Notebook
View Paper

Technologies Used

Python (Pandas, NumPy, Statsmodels, Pmdarima, Scikit-learn, Matplotlib, Seaborn)
Jupyter Notebook for analysis and modeling
Interactive Plotly dashboard for visualization and exploration

Recommendations

Deploy SARIMAX forecasting to improve real-time retail demand prediction.
Integrate external features (weather, holidays, pricing) to improve accuracy.
Use forecasts to align marketing and inventory decisions:
- Plan promotions during forecasted high-demand periods.
- Adjust stock levels proactively to prevent shortages or overstocking.

Risks and Ethical Considerations

Data Limitations: Synthetic datasets may not fully reflect real-world behavior.
Ethics: Only non-sensitive, publicly available data were used.
Mitigation: Retrain models periodically using fresh and validated datasets.