3612. Using machine learning for constraint learning in multiproduct pricing optimization
Invited abstract in session MC-59: Pricing and learning 2, stream Pricing and Revenue Management.
Monday, 12:30-14:00Room: S08 (building: 101)
Authors (first author is the speaker)
| 1. | Luis Aburto
|
| Facultad de Ingeniería y Ciencias, Universidad Adolfo Ibañez | |
| 2. | Vicente Ahumada
|
| Facultad de Ingeniería y Ciencias, Universidad Adolfo Ibáñez |
Abstract
Multiproduct pricing optimization is a challenging process. Calibrating demand models based on transactional data tends to produce bias in cross-elasticity effects, mainly due to endogeneity issues, getting nonsensical or extreme solutions in the optimization process. We explore and compare different supervised and unsupervised methods to extract constraints from the data to formulate a robust feasible space for the pricing optimization procedure. These constraints or pricing rules uncover latent business rules where relationships between product prices happen. Also, they can identify price solutions correlated with good product category performance. We apply our methodology using the orange juice category dataset for different supermarket stores, obtaining robust improvements from 15% to 40% in expected category profit. Some of the methods used to extract pricing rules from data are association rules, SVM classifiers, robust optimization, and ellipsoidal kernels.
Keywords
- Revenue Management and Pricing
- Robust Optimization
- Machine Learning
Status: accepted
Back to the list of papers