Forecasting E‑Commerce Sales with Gradient Boosting
summary
An automatic warehouse restocking system at OZON.ru is driven by a gradient boosting ML model for sales forecasting. The talk covers problem setup and loss-function choice, and describes feature engineering (about 180 features) including how features were created and selected to capture complex seasonality, competitor entry, sudden viral spikes and drops.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 2:52Why forecast sales
- 6:04Choosing metrics and goals
- 7:50Simple models and first results
- 9:34Feature engineering and domain knowledge
- 15:51Training data challenges
- 19:03Model training and early improvements
- 21:51Hyperparameter tuning issues
- 24:38Model tuning and regularization
- 26:28Results and model evaluation
- 27:56Forecast integration
- 31:51Automated purchasing and pipeline
- 34:00Model monitoring and maintenance
- 37:31Factor analysis
- 40:18Model development plans
- 43:27New product forecasting
- 51:17Business impact and ROI
- 53:43Overfitting and future tasks
- 54:03Q&A
related talks
31mFrom stockouts to happy customers: Proven solutions for time series forecasting in retail
09mDemand Forecasting in Supply Chain Using AI Basis Machine Learning
56mAutomatic Revenue Forecasting with Boosting Models and DL
28mMastering Demand Forecasting: Lessons from Europe's Largest Retailer
23mMachine Learning with a Deep Dive into Automated Forecasting System
59mPredictability Check: Demand Forecasting
30mSupporting B2Bsales forecasting by machine learning
08mMachine Learning for Bangladeshi Retailers