Store Sales Forecasting
Domain: Retail Operations
Streamlit Live Web Interface for Real-Time Prediction.
Business Problem:
A drugstore chain needed accurate 6-week sales forecasts to optimize inventory and prevent stockouts/overstocking across 1,115 stores.
My Approach:
o Conducted Time Series EDA to uncover weekly seasonality (Sunday closures) and the financial impact of promotions.
o Engineered features for Seasonality (DayOfWeek, Month) and Business Events (Promo, State Holiday).
o Refinement: Identified and fixed a critical Data Leakage issue (removing the ‘Customers’ future variable) to ensure the model was realistic for production use.
Key Results:
o Quantified that Promotions drive a median sales lift of ~$3,000 per day.
o Built a Random Forest Regressor achieving a robust R² of ~0.85 (MAE ~$600-800) after removing data leakage.
o Accurately predicted sales spikes and “Closed Day” drops in the validation set.