Detection of amylose content in rice samples with spectral augmentation and advanced machine learning

Amylose content (AC) in rice samples constitutes one of the many essential indicators for determining its eating taste and quality. This paper describes an integrated strategy using spectroscopy (12000–4000cm−1), spectral pre-processing, data/spectral augmentation, wavelengths/features selection, and sophisticated machine learning (ML) approaches to identify AC in rice samples reliably. Two scenarios were tested: one with the original dataset and another with SMOTE-augmented spectral dataset. Furthermore, the performance of the models was also evaluated on segmented wavelengths range 9000–4000 cm−1. The most informative wavelengths were chosen and used in different ML models, including stacking regressors and classifiers for forecasting the AC. The regression analysis yielded the best results with a coefficient of determination (R²) of 0.968, Root Mean Squared Error (RMSE) of 0.306, and Ratio of performance of deviation (RPD) of 5.564. The classification performance achieved an F1 score of 99.0 %, with a 5-fold average F1 score of 91.2 %. These findings show that the suggested technique can identify AC in rice in a non-destructive and dependable manner, significantly improving model performance over previous works.