Classification of Iranian rice varieties using FTIR spectroscopy and sparse Linear Discriminant Analysis

Authentication and classification of rice varieties have attracted much attention in food science and technology for quality assurance and adulteration detection in recent years. In this study, FTIR spectroscopy was integrated with chemometrics techniques for discrimination of seven different ‘Iranian rice’ varieties. Herein, for the first time, the sparse version of Linear Discriminant Analysis (sLDA) was used for the development of the interpretable and reliable FTIR-based classification model to achieve the discrimination of rice samples. The development of the sLDA algorithm was done by applying an elastic net penalty to the discriminant vectors in the optimal scoring explanation of LDA, which could simultaneously perform classification and variable selection. A grid search technique combined with 5-fold cross-validation was used to tune the parameters and optimizing the model. The performance of the sLDA model was compared with LDA, and the accuracy of the validation set increased from 89.52% for the LDA model to 95.24% for the sLDA model. The results indicated that the sLDA outperformed the LDA in terms of interpretability and prediction accuracy, which is mostly due to sparse loading vectors in the sLDA method. The present work revealed that a tuned and optimized sLDA technique combined with FTIR data can be successfully applied as a robust and interpretable tool for authentication and classification of the rice samples in food industry [1, 2].