Previsão de demanda através de redes neurais em um ambiente de omnicanalidade no varejo

Abstract

Given the retail development towards an omnichannel environment and the consequent increase in available information sources, demand forecasting becomes increasingly complex. Thereby, the search for solutions that adapt to this scenario becomes relevant. A literature review analyzed the main demand prediction models using artificial intelligence algorithms from 2010 to 2020, as well as their results and comparisons with traditional models. This work presents a data analysis model that adopts the combination of two machine learning algorithms (hierarchical clustering and artificial neural networks) to forecast demand from training on two-year data from a local retailer in an omnichannel environment containing internal and external attributes. The model demonstrated, by the accuracy measures MAPE, MAE, and RMSE, values consistent with the results obtained in the literature for demand forecasting, despite the omnichannel environment. The insights of this research can be applied in fulfillment and allocation policies in different channels. Therefore, the generalization of this model provides the opportunity for further research, taking into account other attributes, such as the consumer profile, as well as variations of the algorithm to improve its robustness