Modeling of Convective Coffee Drying Using an Artificial Neural Network

Karidioula Daouda *

Department of Sciences and Technologies, Chemistry Laboratory, Man University, BP 20 Man, Côte d’Ivoire and Department of Chemical and Agro-Food Engineering, National Polytechnic Institute in Yamoussoukro, BP 1313, Yamoussoukro, Côte d’Ivoire.

Zran Vanh Eric-Simon

Department of Material Structure Sciences and Technology, Laboratory of Physical Chemistry, Félix Houphouët Boigny University, 22 BP 582 Abidjan 22, Côte d'Ivoire.

Kouakou Lébé Prisca Marie-Sandrine

Department of Material Structure Sciences and Technology, Laboratory of Constitution and Reaction of Matter, Félix Houphouët Boigny University, 22 BP 582 Abidjan 22, Côte d'Ivoire.

Assidjo Nogbou Emmanuel

Department of Chemical and Agro-Food Engineering, National Polytechnic Institute in Yamoussoukro, BP 1313, Yamoussoukro, Côte d’Ivoire.

Trokourey Albert

Department of Material Structure Sciences and Technology, Laboratory of Physical Chemistry, Félix Houphouët Boigny University, 22 BP 582 Abidjan 22, Côte d'Ivoire.

*Author to whom correspondence should be addressed.


Abstract

Convective drying is used in coffee processing, and modelling moisture-content evolution may help describe drying behaviour under controlled conditions. This study assessed the ability of an artificial neural network (ANN) to represent changes in coffee moisture content during laboratory convective drying. Samples weighing approximately 250 g were dried at 40, 60, 80 and 100°C, with three trials at each temperature. Drying time, drying temperature and initial moisture content were used as model inputs; moisture content at the corresponding drying time was the output. A multilayer perceptron with three hidden layers was implemented using Python. Experimental observations were allocated to training (60%), validation (20%) and test (20%) subsets. Fifteen hidden-layer configurations were evaluated using the coefficient of determination (R²) and root mean square error (RMSE). The 3-128-32-64-1 configuration achieved the highest reported test R² (0.99996) and the lowest test RMSE (0.3838) among the configurations considered. For eight randomly selected observations from the experimental dataset, the reported mean absolute errors ranged from 0.0005 to 0.063. The trained model was also used to generate estimates at temperatures not included in the experimental trials, but these predictions were not independently validated. Overall, the ANN reproduced the observed moisture-content patterns under the investigated laboratory conditions. Further experimental evaluation is required to assess its performance under other drying conditions and its suitability for practical applications.

Keywords: Coffee, artificial intelligence, modelling, artificial neural network (ANN), drying


How to Cite

Daouda, Karidioula, Zran Vanh Eric-Simon, Kouakou Lébé Prisca Marie-Sandrine, Assidjo Nogbou Emmanuel, and Trokourey Albert. 2026. “Modeling of Convective Coffee Drying Using an Artificial Neural Network”. Asian Journal of Chemical Sciences 16 (6):87-95. https://doi.org/10.9734/ajocs/2026/v16i6493.

Downloads

Download data is not yet available.