Comparison of neural networks and support vector machines for predicting the ripening time of Swiss type cheese: a dielectric spectroscopy application

Castro W.; Cieza Y.; Castro-Giraldez M.; Fito P.J.; Chuquizuta T. · Applications in Software Engineering - Proceedings of the 14th International Conference on Software Process Improvement, CIMPS 2025 (2025) · DOI: 10.1109/cimps69356.2025.11419695

Resumen

Cheese ripening, due to the biochemical changes it generates, impacts quality and marketability; therefore, the development of evaluation methods is of interest. Indeed, noninvasive techniques, such as dielectric spectroscopy, are of particular interest, especially when coupled with chemometric techniques, and their further study is necessary. Therefore, it was proposed to compare models based on artificial neural networks (ANN) and support vector machines (SVM) coupled with dielectric profiles to predict cheese ripening time. For this purpose, vacuum-packed Swiss cheese samples were stored at 10°C for ripening. Their dielectric profiles were extracted every 15 days in the radiofrequency range of [0.04 - 1.00] MHz. The profiles were preprocessed to express them in terms of e′ and”e””; to build the complete regression models based on ANN and SVM, the relevant variables were first identified. With these variables, optimized models were then developed. All models were trained by splitting the data as follows: 70 % for calibration and 30 % for validation. This process was repeated thirty times, and in each iteration, the corresponding statistical metrics (R2, RMSE) were calculated. The implemented models achieved, on average, R2 values greater than 80 % and RMSE below 10. The results show the feasibility of predicting the ripening time of Swiss cheeses using dielectric spectroscopy coupled with chemometric tools; likewise, for this case, ANN were slightly superior to SVM.

Instituciones: Universidad Nacional de Trujillo; Universidad Nacional Autónoma de Chota; Universidad Nacional de Frontera