NIRS and machine learning algorithms as a non-invasive technique to discriminate and classify cooked broiler and duck meat
This study investigated the utilization of near-infrared spectroscopy (NIRS) and machine learning methodologies to differentiate cooked broiler and duck meat. Nearinfrared spectral data were acquired using a portable spectrometer (700–1100 nm) from 40 samples (20 broilers and 20 duck breasts). The d...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-06-01
|
| Series: | Applied Food Research |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772502225002926 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study investigated the utilization of near-infrared spectroscopy (NIRS) and machine learning methodologies to differentiate cooked broiler and duck meat. Nearinfrared spectral data were acquired using a portable spectrometer (700–1100 nm) from 40 samples (20 broilers and 20 duck breasts). The data were preprocessed and analyzed using principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE). PCA revealed 87.97 % variance, demonstrating distinct groupings. Various machine learning algorithms, including linear discriminant analysis (LDA), support vector machine (SVM), Logistic Regression, Random Forest, Gradient Boosting, K-Nearest Neighbors, Decision Tree, Naive Bayes, Neural Network, and XGB, were evaluated. Hyperparameter tuning and 5-fold cross-validation were implemented to mitigate overfitting and enhance the generalization. Logistic Regression, SVM, and LDA achieved the highest accuracy (84.13 %, 84 %, and 83.50 %, respectively) and ROC AUC (92.99 %, 92.63 %, and 91.71 %, respectively), indicating robust discriminative power. Random Forest and Gradient Boosting also exhibited satisfactory performance, with consistent cross-validation scores (∼0.824). XGB demonstrated marginally superior recall compared to K-Nearest Neighbors, while Neural Network excelled in recall (90.74 %), a critical factor for detecting true positives. Decision Tree and Naive Bayes, although interpretable, underperformed relative to other algorithms. Logistic Regression and SVM have emerged as the most efficacious methods, attaining high accuracy and balanced precision and recall. This study underscores the potential of integrating spectroscopic data with advanced machinelearning techniques for meat classification and fraud detection. |
|---|---|
| ISSN: | 2772-5022 |