Predicting the accuracy of regression models in the retail industry

• Type: Article in International Conference Proceedings Book
• Date: 2012

Authors

Pinto, F (Author) Other; Carlos Soares (Author) FEP

Conference proceedings International

Title: CEUR Workshop Proceedings

Pages: 28-29

5th Planning to Learn Workshop, PlanLearn 2012 at 2012 European Conference on Artificial Intelligence, ECAI 2012

28 August 2012

Indexing

Scopus - 0 Citations

Other information

Authenticus ID: P-00K-SPW

Abstract (EN): Companies are moving from developing a single model for a problem (e.g., a regression model to predict general sales) to developing several models for sub-problems of the original problem (e.g., regression models to predict sales of each of its product categories). Given the similarity between the sub-problems, the process of model development should not be independent. Information should be shared between processes. Different approaches can be used for that purpose, including metalearning (MtL) and transfer learning. In this work, we use MtL to predict the performance of a model based on the performance of models that were previously developed. Given that the sub-problems are related (e.g., the schemas of the data are the same), domain knowledge is used to develop the metafeatures that characterize them. The approach is applied to the development of models to predict sales of different product categories in a retail company from Portugal.

Language: English

Type (Professor's evaluation): Scientific