Szczegóły publikacji
Opis bibliograficzny
Deep neural networks in profiling of apple juice adulteration based on voltammetric signal of the iridium quadruple-disk electrode / Szymon WÓJCIK, Małgorzata JAKUBOWSKA // Chemometrics and Intelligent Laboratory Systems ; ISSN 0169-7439. — 2021 — vol. 209 art. no. 104246, s. 1–10. — Bibliogr. s. 9–10, Abstr. — Publikacja dostępna online od: 2021-01-08
Autorzy (2)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 132335 |
|---|---|
| Data dodania do BaDAP | 2021-02-02 |
| Tekst źródłowy | URL |
| DOI | 10.1016/j.chemolab.2021.104246 |
| Rok publikacji | 2021 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Czasopismo/seria | Chemometrics and Intelligent Laboratory Systems |
Abstract
In this work, the recurrent deep neural networks for estimation of the glucose - fructose syrup (GFS) content, deliberately added to apple juices, was described. The signal source was voltammetric measurements performed using the iridium quadruple - disk electrode. Samples of 17 apple juices of various origin were divided into 6 groups, the additions of GFS, up to 50% content, were tested. Considering the set of all juices simultaneously, after adding 0–50% of GFS it was shown, that using the linear multivariate regression methods, the root-mean-square error of prediction (RMSEP) was ca. 6.3–10.8% of GFS. R2 for the prediction set was not higher than 0.86. Finally, in the most optimal approach, using the recurrent deep neural networks with Long Short-Term Memory layer, the RMSEP was 1.9–2.1% of GFS. It is a value ca. three times lower in comparison to the linear calibration models. In this case R2 for the prediction set was greater than 0.98. Application of the deep learning approach has allowed development of a universal model for quantification of the syrup content in apple juice.
