Advanced
Open
Pro
Pick the metric and method for 80,000 slow-moving spare-part SKUs
Part of the ML System Design Interview path →
Part of the Advanced Data Science & Experimentation path →
A parts distributor stocks 80,000 spare-part SKUs. 75 to 85% of daily SKU-store observations are zero, non-zero days sell 1 to 4 units, replenishment lead time is 28 days, and some parts are being phased out. Management reports MAPE and wants "the most accurate model".
- Why is MAPE unusable here, and what does the all-zeros forecast do to MAE and WAPE?
- Name the metrics you would report, and how each one maps to a decision.
- Compare Croston, SBA, TSB and a Tweedie GBM: what does each assume and where does each fail?
- How do you classify the SKUs before choosing a method?
- The business wants a safety-stock level. What changes in the model and in the evaluation?
Share this question