A Predictive Sales System Based on Deep Learning

• Autores: Luyo Ballena, Jean Paul, Ortiz Pallihuanca, Cristhian Pool, Carrera Salas, Ernesto Adolfo
• Formato: artículo
• Fecha de Publicación: 2024
• Institución: Universidad Peruana de Ciencias Aplicadas
• Repositorio: UPC-Institucional
• Lenguaje: inglés
• OAI Identifier: oai:repositorioacademico.upc.edu.pe:10757/675749
• Enlace del recurso: https://doi.org/10.14569/IJACSA.2024.0150117; http://hdl.handle.net/10757/675749
• Nivel de acceso: acceso abierto
• Materia: Deep learning; neural network architectures; neural networks; sales prediction; https://purl.org/pe-repo/ocde/ford#3.00.00

There are several techniques for predictive sales systems, in this study, a system based on different machine learning algorithms is developed for a trading company in Lima. As any company, it needs to be accurate in its sales calculations to manage the volume of production or product purchases. With the system, the trading company has a mechanism to order products from its supplier based on the predictions and estimates of the needs according to the projection of its sales. For the sales predictive system, Deep Learning technology and the neural network architectures GRU (Gated Recurrent Unit), LSTM (Long Short Term Memory) and RNN (Recurrent Neural Network) were used, 10 products were sampled, and the sales quantities of the last 12 months were obtained for the evaluation. The study found that the LSTM architecture excels in accuracy, significantly outperforming GRU and RNN in terms of Mean Absolute Percentage Error (MAPE), achieving an average MAPE of 7.07%, in contrast to the MAPE of 27.14% for GRU and the MAPE of 36.17% for RNN. These findings support the effectiveness and versatility of LSTM in time series prediction, demonstrating its usefulness in a variety of real-world applications.