Comparison Analysis of FB Prophet, Amazon DeepAR+ & CNN-QR Algorithms
summary
Real-world retail sales from a 5,081‑item portfolio (400 active, top 200 ≈90% turnover) are backtested 12 times monthly to compare Facebook Prophet, Amazon DeepAR+ and CNN‑QR; Prophet delivers strong long‑history results (example item MAPE ≈8% monthly, ≈6% quarterly) and about 40% of SKUs hit MAPE<15% quarterly covering ~80–90% financial share, while DeepAR+/CNN‑QR beat Prophet on short histories and a per‑item voting ensemble chooses the best model; the pipeline runs in production and sustains service levels during COVID while cutting inventory from the black to the green line.
AI-written summary · about
chapters (7)
- 0:00Start
- 4:16Real scenarios
- 11:31Data Filtering and Preprocessing
- 12:35Prophet - a Tool for Time Series Forecasting at Scale
- 14:14Performance Metrics
- 15:04Forecastability Analysis using Backtesting Experiments
- 24:53Sales Forecasting - Results - COVID analysis
related talks
41mNeural Prophet – A powerful AI framework for Time Series Models
35mBuilding a successful inventory planning solution with Amazon Forecast
44mRobust AI Search Ranking for Radical C2C Marketplace Growth
27mOptimizing Omnichannel Order Fulfillment with AI and Advanced Analytics
28mBuild Facebook's Prophet in PyMC3
40mMulti-Series Forecasting at Scale with StatsForecast
35mCan Memes Predict Markets? Using Social Media Sentiment for Stock Forecasting
37mAI/ML Strategies & Techniques for a Post Pandemic World