Feasibility of fraud detection in rice using a handheld near-infrared spectroscopy

Food fraud continues to be a global issue. In recent years, fraudulent rice draws attention to the public agency where rice substituted with look-alike substance, low-quality rice, or impure substances due to profit motive. However, the techniques used to determine fraudulent rice, such as DNA profiling and physicochemical properties, are laborious, inconvenient, and time-consuming. The Near-Infrared Spectroscopy (NIRS) increased acceptance in fraud detection in recent years due to its good feasibility, accuracy and non-destructive. This study utilizes NIRS, Principal Component Analysis (PCA), and Logistic Regression (LR) to explore the correlated variable and to determine the linear relationship between the spectral of adulteration of rice sample. A total of 123 near infrared (NIR) spectral data collected from 31 unadulterated rice samples and ten adulterated rice samples in 3 different lightning condition places. Based on the processed data in PCA, the LR model achieved good accuracy of 94.4% on training and 99.4% on the independent test set. This study indicated that the combination of NIRS, PCA, and LR is feasible and effective in fraud detection in rice.