AI & Machine Learning / Interactive Demo

Demand Forecasting & Inventory Decision Support

Weekly multi-SKU forecasts turned into inventory risk and replenishment signals — with the uncertainty kept visible.

Time Series Explainable ML Inventory Planning Python
12 SKUs156 weekly observations each
8 windowsRolling-origin · final 8 weeks isolated
Seasonal NaiveBest aggregate WAPE on the frozen test

Synthetic weekly demand · chronological validation · one frozen test period

The more complex model did not automatically win.

Seasonal Naive produced the best aggregate WAPE on the frozen test period; the validation-selected combination did not generalize better. The result is retained rather than tuned away.

Candidates
Seasonal Naive · damped ETS · global gradient boosting
Selection
Per SKU, by validation error with a simplicity tie-break
Decision rule
Weekly review · lead-time demand, safety stock, reorder point
  • 12-SKU scenario, or upload a validated weekly CSV
  • Triage SKUs by current stockout risk
  • Forecast history, uncertainty ranges, backtest metrics
  • Change lead time, service level, demand growth
  • Trace any recommendation back to its calculation

On-hand inventory only — no open purchase orders. No holding cost, shelf life, order minimums, supplier capacity, or ERP execution. Forecast ranges summarize observed backtest error, not future demand.