Arabica coffee price forecast: a neural network application CNN-BLSTM

This work proposes the use of the CNN-BLSTM neural network as a tool to predict the price of arabica coffee. The database provided by CEPEA (Center for Advanced Studies in Applied Economics) presents a historical series of the price of arabica coffee, in the period between January 1997 and December...

Full description

Bibliographic Details
Author: Santos, José Airton Azevedo dos
Format: article
Status:Published version
Publication Date:2022
Country:Brasil
Institution:Universidade Federal de Itajubá (UNIFEI)
Repository:Research, Society and Development
Language:Portuguese
OAI Identifier:oai:ojs.pkp.sfu.ca:article/26101
Online Access:https://rsdjournal.org/index.php/rsd/article/view/26101
Access Level:Open access
Keyword:Artificial neural networks
Arabica coffee
Keras
Python.
Redes neuronales artificiales
Café arábica
Redes neurais artificiais
Description
Summary:This work proposes the use of the CNN-BLSTM neural network as a tool to predict the price of arabica coffee. The database provided by CEPEA (Center for Advanced Studies in Applied Economics) presents a historical series of the price of arabica coffee, in the period between January 1997 and December 2021. Forecast models based on neural networks LSTM, BLSTM, CNN and CNN-BLSTM were implemented, in the Python language, using the Keras framework. Results obtained, from the four models, were compared using MAE, RMSE and MAPE metrics. It was verified, for a horizon of 6 months, that the CNN-BLSTM model presented better performance.