Applying LSTM Recurrent Neural Networks to Predict Revenue

Autor(es)

Pelin Cardoso, Luis Eduardo ; de Carvalho, Andre C. P. de Leon F. ; Quiles, Marcos G.

• Tipo de documento: Artigo Científico
• Fonte: COMPUTATIONAL SCIENCE AND ITS APPLICATIONS-ICCSA 2024, PT II; v. 14814, p. 15-pg., 2024-01-01.

Resumo

Predicting a company's revenue is not a trivial task. Attempting to identify a sales pattern for a store's products, considering all possible variables, both internal and external to the business, is a complex task. For companies that sell products, finding a way to predict next day's revenue helps maintain lean inventory, among numerous other factors. As result, fewer resources are wasted in the form of stagnant products, and there are less expenses in purchasing supplies. The main motivation for this study is to accurately forecast a company's revenue for the days, following the prediction day-only for companies that sell products and disregarding those that offer services. To achieve this goal, here, considering a real scenario, we evaluate the use of the Long Short-Term Memory (LSTM) type of Recurrent Neural Network, known for their superior ability to capture long-term temporal dependencies, outperforming traditional forecasting models in this domain. For our experiments, we employ a single real data model across various LSTM architectures, iteratively refining our approach to approximate an ideal state for future revenue prediction. This study not only validates the effectiveness of LSTM networks in handling the temporal complexity inherent in sales data but also investigates the impact of different architectural configurations on forecasting accuracy. By doing so, we aim to identify a robust model that can serve as a reliable tool for businesses in their inventory and supply chain management strategies. (AU)

• Processo FAPESP: 20/09835-1 - IARA - Inteligência Artificial Recriando Ambientes
• Beneficiário: André Carlos Ponce de Leon Ferreira de Carvalho
• Modalidade de apoio: Auxílio à Pesquisa - Programa Centros de Pesquisa em Engenharia
• Processo FAPESP: 22/09285-7 - Exploração do espaço químico via aprendizado semissupervisionado para geração de novos materiais
• Beneficiário: Marcos Gonçalves Quiles
• Modalidade de apoio: Auxílio à Pesquisa - Regular