Gestão preditiva dos custos de alimentação de suínos utilizando redes neurais híbridas CNN-LSTM

Authors

DOI:

https://doi.org/10.22279/navus.v18.2060

Abstract

Feed cost is the factor that most influences the total cost of pig production. For the producer, its forecast is fundamental in managing his business. In this context, this work proposes using the multivariate CNN-LSTM (Long Short Term Memory - Convolutional Neural Network) network to forecast the feed cost for pig feeding in the state of Paraná. The database is formed, in the period from January/2007 to May/2024, by monthly series of the cost of feed, the price of soybean meal and the price of corn. Univariate and multivariate forecasting models were implemented in Python using the Keras framework. Results of CNN-LSTM and LSTM forecasting models, in their univariate and multivariate versions, were compared using the MAE (Mean Absolute Percent Error), RMSE (Root Mean Squared Error), and MAPE (Mean Absolute Percent Error) metrics. It was found that the multivariate CNN-LSTM model, for a 12-month horizon, presented the best performance (MAE=0.068, RMSE=0.011 and MAPE=1.70).

Downloads

Download data is not yet available.

Author Biographies

Aldino Normelio Brun Polo, Universidade Tecnológica Federal do Paraná

Mestrando do PPTGCA: Programa de Pós-Graduação em Tecnologias Computacionais para o Agronegócio

José Airton Azevedo dos Santos, Universidade Tecnológica Federal do Paraná

Programa de Pós-Graduação em Tecnologias Computacionais para o Agronegócio (PPGTCA)

Published

2026-08-24

Issue

Section

Articles