Use this link to cite:
https://hdl.handle.net/2183/49485 Desarrollo de un modelo predictivo de ventas basado en segmentación avanzada de un catálogo de prendas de moda
Loading...
Identifiers
Publication date
Authors
Fernández Pan, Mauro
Advisors
Other responsabilities
Universidade da Coruña. Facultade de Informática
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen] El sector del retail de moda se caracteriza por una alta volatilidad de la demanda y cadenas de suministro con tiempos de fabricación rígidos (de hasta 4 meses), lo que genera riesgos financieros críticos como las roturas de stock o el exceso de inventario inmovilizado. El prin cipal reto analítico de este proyecto reside en el problema del cold-start —la predicción para artículos nuevos sin historial— y en el sesgo provocado por las roturas de stock históricas.
Este trabajo propone el diseño e implementación de una arquitectura predictiva avanzada estructurada en dos fases. La primera consiste en un sistema de segmentación del catálogo mediante aprendizaje no supervisado que agrupa los artículos según sus similitudes físicas y de negocio. La segunda fase utiliza esta segmentación como base para un motor de pronóstico mediante modelos de gradient boosting. Con un enfoque de predicción jerárquica, el sistema permite inferir la demanda de nuevos productos a partir del comportamiento de su clúster, optimizando la planificación logística y protegiendo los márgenes comerciales.
[Abstract] The fashion retail sector is characterized by high demand volatility and rigid supply chains with long lead times (up to 4 months), leading to critical financial risks such as stockouts or excessive immobilized inventory. The main analytical challenge of this project lies in the cold start problem —forecasting sales for new items without historical data— and the bias caused by previous stockouts. This project proposes the design and implementation of an advanced predictive architecture structured in two phases. The first phase consists of a catalog segmentation system using unsupervised learning that groups items based on their physical and business similarities. The second pase uses this segmentation as a foundation for a forecasting engine based on gradient boosting models. Using a hierarchical forecasting approach, the system allows inferring the demand for new products based on the behavior of their corresponding cluster, thus optimizing logistical planning and protecting commercial margins.
[Abstract] The fashion retail sector is characterized by high demand volatility and rigid supply chains with long lead times (up to 4 months), leading to critical financial risks such as stockouts or excessive immobilized inventory. The main analytical challenge of this project lies in the cold start problem —forecasting sales for new items without historical data— and the bias caused by previous stockouts. This project proposes the design and implementation of an advanced predictive architecture structured in two phases. The first phase consists of a catalog segmentation system using unsupervised learning that groups items based on their physical and business similarities. The second pase uses this segmentation as a foundation for a forecasting engine based on gradient boosting models. Using a hierarchical forecasting approach, the system allows inferring the demand for new products based on the behavior of their corresponding cluster, thus optimizing logistical planning and protecting commercial margins.
Description
Editor version
Rights
Todos os dereitos reservados. Prohíbese a reprodución, transformación, distribución e comunicación pública da obra por terceiros. Permítese a súa visualización e a descarga dunha copia privada para o uso persoal.






