Optimizing DD-SIMCA Modeling for Accurate Classification of Rice Varieties via Raman Spectroscopy

Rice variety significantly affects its processing and quality This study established an efficient, fast, trustworthy and nondestructive method for identifying rice varieties by coupling Raman spectroscopy and multivariate data analysis. 164 rice samples, consisting of 68 Hashemi, 21 Tarom, 24 Fajr, and 51 Shirudi, were acquired. Raman spectra were obtained from these samples in the 431–3470 cm-1 spectral range. The rice samples were then separated into training and validation sets using the Kennard-Stone (70–30) algorithm. The smoothing and differential pretreatment algorithms were applied to prepare the spectra for analysis. Raman spectroscopy and data-driven soft independent modeling of class analogy 10 (DD-SIMCA) were employed. Individual modeling sets for Hashemi and Tarom samples were used to establish the DD-SIMCA classification models based on principal component analysis (PCA). The DD-SIMCA modeling set, comprising two single-class classifiers, identifies and differentiates these valuable Iranian rice varieties. The models developed for the classification of the two types of rice show 100% sensitivity at a 95% confidence level, while the specificity of the models for Hashemi rice ranged from 85 to 100%, and, for Tarom rice, it ranged from 8 to 100%. The trial results clearly show that the DD-SIMCA model is highly efficient in distinguishing preferred rice varieties from adulterated samples, even when mixed with non-valuable types of rice.