Comparison of recurrent neural networks and partial least squares regression for predicting coffee quality using Near Infrared spectroscopy

Castro W.; Juarez L.; Tene B.; Gonzales J.; Berru J.; Acevedo-Juarez B.; Avila-George H. · Applications in Software Engineering - Proceedings of the 12th International Conference on Software Process Improvement, CIMPS 2023 (2023) · DOI: 10.1109/cimps61323.2023.10528851

Resumen

Coffee is an important agricultural product, and its quality depends on various factors. Cupping is subjective and expensive, so researchers have sought faster and more reliable techniques. NIR spectroscopy has shown promising results in evaluating coffee quality. This study compares the prediction capabilities of recurrent neural networks (RNN) and partial least squares regressions (PLSR) for predicting coffee quality. Six samples of ground coffee were used, and NIR spectral profiles were obtained. Prediction models were constructed using PLSR and RNN, and the prediction capabilities of both models were evaluated. Relevant variables were selected to optimize the models, and performance metrics were calculated. The results of this study can contribute to the development of faster and more reliable methods for assessing coffee quality, benefiting the coffee industry in terms of efficiency and product quality.

Instituciones: Universidad Nacional de Frontera