Intermittent Demand Forecasting on M5 (Walmart)
Walmart products often sell sparsely — many days with zero sales, occasional bursts. Standard forecasting fails on this; this app compares classical statistical methods, machine learning, and deep learning on real (anonymized) Walmart product-store demand.
How to use: Pick a product below — the label tells you its demand pattern (e.g. sells rarely, spiky sizes), since Walmart's real product names are anonymized in this public dataset. Watch how methods behave differently on smooth vs. sparse items. The Stocking decision tab turns a forecast into an inventory recommendation in ₹.
Selected: Food item in store WI_1 (WI). Demand class: Erratic. Zero-sales on ~21% of days; ADI=1.27, CV²=0.71. Higher ADI = sells less often; higher CV² = spikier sizes.
Accuracy on this SKU (sorted by RMSSE)
Inventory cost over the 28-day horizon
Costs are illustrative assumptions. Holdout = M5 validation window (d_1914–d_1941); not directly comparable to the official M5 leaderboard.