Ensemble learning from model based trees with application to differential price sensitivity assessment
The assessment of price sensitivity is a relevant issue with important implications in decision making for revenue management. The issue has attracted interest among companies evolving towards the data-driven culture through the exploitation of their data sources. Thus, the design of pricing strateg...
| Autor: | |
|---|---|
| Tipo de documento: | artigo |
| Data de publicação: | 2021 |
| País: | España |
| Recursos: | Universidad Nacional de Educación a Distancia |
| Repositório: | e-spacio. Repositorio Institucional de la UNED |
| Idioma: | inglês |
| OAI Identifier: | oai:e-spacio.uned.es:20.500.14468/26322 |
| Acesso em linha: | https://hdl.handle.net/20.500.14468/26322 |
| Access Level: | Acceso aberto |
| Palavra-chave: | ensemble learning differential price sensitivity model based recursive partitioning revenue management |
| Resumo: | The assessment of price sensitivity is a relevant issue with important implications in decision making for revenue management. The issue has attracted interest among companies evolving towards the data-driven culture through the exploitation of their data sources. Thus, the design of pricing strategies that rely on analytics to identify groups of customers that exhibit differential price sensitivity has a great potential for revenue managers. This work proposes a data-driven approach, using ensemble learning from model based trees, to assess differential price sensitivity in a similar way as random forests algorithm does to assess variable importance. A differential price sensitivity score is defined and a ranking is obtained as a result so that the top ranked variables can be selected as candidate inputs for segmentation and differential price sensitivity group finding. Then optimal price allocation is carried out on the derived groups in order to compute the expected revenues which are compared with the revenues given by un-optimized prices and by optimal price allocation derived from the logit estimation of the bid response function. The proposed approach is validated in synthetic experiments and by application to the real business case of an auto lending company; the resulting revenues show its benefit. |
|---|