Part 5/11:

• Demand Forecasting: Forecasting algorithms predict future sales over various horizons, from days to months. Given data nuances, the system employs different algorithms, such as Exponential Moving Averages, XGBoost, Elastic Net, and LSTM models. The models are trained monthly, with weekly hyperparameter tuning to adapt to changing patterns.

• Segmentation: Items are classified into categories like high velocity (Gold), moderate (Silver), or low (Bronze) to improve forecast accuracy tailored to their sales frequency and value.

• Granular Alerts: Since direct real-time shelf monitoring isn't feasible everywhere, the system extrapolates from weekly or monthly data into daily forecasts, ensuring timely alerts that align with daily retail operations.