Enhancing soluble dietary fiber prediction in barley via Boruta-based feature selection and mid-infrared spectroscopy

Barley (Hordeum vulgare L.) is a major cereal crop whose soluble dietary fiber (SDF) offers significant health benefits, yet conventional SDF determination methods are time-consuming, labor-intensive, and destructive to samples. This study developed a rapid, non-destructive method for SDF quantification in barley using mid-infrared (MIR) spectroscopy combined with a Boruta-based partial least squares (PLS) hybrid approach. A total of 280 barley grain samples were subjected to Fourier-transform infrared (FTIR) spectral acquisition from 2000 to 650 cm^-1, with reference SDF values determined by the association of official analytical chemists (AOAC) 991.43 enzymatic-gravimetric method. The Boruta algorithm selected 165 informative wavenumbers out of 363, reducing the variable space by 54.5%. The developed Boruta-PLS model achieved excellent predictive performance with a coefficient of determination for the training set (R²_C) of 0.9695 and for the test set (R²_P) of 0.9601 and a root mean squared error for the training set (RMSE_C) of 0.7464%, and for the test set (RMSE_P) of 0.7340%, substantially outperforming the full-spectrum PLS model with an R²_P of 0.7759 and an RMSE_P of 1.7387%, as well as conventional wavelength selection methods including variable importance in projection (VIP), competitive adaptive reweighted sampling (CARS), and uninformative variable elimination (UVE). The selected wavenumbers were predominantly located in chemically relevant regions at 1000-1200 cm^-1 and 1500-1700 cm^-1, confirming model interpretability. This Boruta-PLS approach provides a rapid, non-destructive, and cost-effective alternative for SDF quantification, with strong potential for high-throughput screening and real-time quality monitoring of barley samples.