Designating the geographical origin of Iranian almond and red jujube oils using fluorescence spectroscopy and l1-penalized chemometric methods
•Almond oils and red jujube oils extracted from plants of mountain areas in Iran were correctly classified.•sNPLS-DA algorithm was used for analysis of EEM fluorescence data for the first time.•The CCD approach was used for swift optimization of sNPLS-DA models.•The sNPLS-DA models were easily inter...
Gespeichert in:
Veröffentlicht in: | Microchemical journal 2020-09, Vol.157, p.104984, Article 104984 |
---|---|
Hauptverfasser: | , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | •Almond oils and red jujube oils extracted from plants of mountain areas in Iran were correctly classified.•sNPLS-DA algorithm was used for analysis of EEM fluorescence data for the first time.•The CCD approach was used for swift optimization of sNPLS-DA models.•The sNPLS-DA models were easily interpretable due to sparse loading coefficients.•The performances of sNPLS-DA models were superior over NPLS-DA algorithm.
Assessment of the geographical origin of medical plants (MP) has received a lot of attention in traditional medicine for determination of the quality of the products and detection of food/drug adulteration. In the present contribution, the sparse version of N-way partial least square discriminant analysis (sNPLS-DA) has been used for classification of almond and red jujube oil samples using their excitation-emission (EEM) fluorescence spectra, for the first time. Seven types of almond seeds and six species of red jujube samples were collected from mountains and hills in Fars and Khorasan provinces in Iran, respectively. The sNPLS-DA algorithm with l1-norm constraint was utilized for calculation of sparse loading vectors. This method merges variable selection and modeling steps into one phase, and due to sparse loading vectors, models are easily interpretable. The number of latent variables and non-zero elements in the first and second loading spaces for the sNPLS-DA models in this work was optimized using a central composite design (CCD) approach. The optimized models were used for classification of almond and red jujube oil samples according to their geographical origins in Iran. The average accuracies of the sNPLS-DA models were 0.95 and 0.89 for bootstrapping cross validation and 0.97 and 1.00 for external validation for almond and red jujube samples, respectively. The performance of the sNPLS-DA models was compared with PLS-DA, parallel factor analysis (PARAFAC), counter propagation artificial neural networks (CPANN), and supervised Kohonen networks (SKN). The results of the optimized sNPLS-DA models were superior to PARAFAC-DA and comparable to those of CPANN and SKN models. The results in this work reveal that the sNPLS-DA method can be used as a reliable and interpretable tool for quality control and classification of food samples using their EEM fluorescence data. |
---|---|
ISSN: | 0026-265X 1095-9149 |
DOI: | 10.1016/j.microc.2020.104984 |