Forecast Model Leaderboard
The ensemble wins on most series (wMAPE ~11%), but on low-volume make-to-order electronics a simple regression beats it — the smarter call there.
Best model
Ensemble
wMAPE 11%
Lift vs naive
+16 pts
+16%accuracy gain
Confidence
81%
+3%plan-weighted
Model accuracy (lower is better)
Weighted MAPE across all series
AI narrative
Why this matters
The ensemble leads overall (wMAPE ~11%) by blending Holt-Winters, SARIMA, Prophet-style and regression learners with backtested weights. On Smart Electronics, two NPI items break the seasonal pattern, so a regression model using the sales-pipeline driver cuts error by ~8 points — switch those 14 items and flag for planner review.