Least absolute shrinkage and selection operator-selected NIR reflectance discrete features for robust glucomannan content estimation in Amorphophallus muelleri Blume flour

Glucomannan is a critical quality parameter of Amorphophallus muelleri Blume (porang) flour and is commonly quantified using wet chemical analysis with 3,5-dinitrosalicylic acid (DNS). However, this method is labor-intensive and highly dependent on careful sample preparation, limiting its suitability for rapid, large-scale, real-time, and on-line applications. To address these constraints, we developed multivariate calibration models based on near-infrared (NIR) reflectance spectra of porang flour collected over 1000–2500 nm. Principal component analysis (PCA) was first used to explore spectral variation and revealed clear separation between samples with low-to-moderate (< 50) and high (> 50) glucomannan contents across the first three principal components. A third-degree univariate polynomial regression provided moderate prediction performance R2p of 0.773 and RMSEp of 3.601 but showed reduced reliability for samples below 36 glucomannan. Therefore, partial least-squares regression (PLSR), Gaussian process regression (GPR), and support vector regression (SVR) were evaluated using both full spectra and twelve discrete wavelengths selected via least absolute shrinkage and selection operator (LASSO), combined with spectral preprocessing. The LASSO-selected wavelengths with standard normal variate (SNV) preprocessing and GPR achieved strong accuracy (R2p = 0.975; RMSEp = 1.184; RPD of 6.388), which is comparable to the full-spectrum model (R2p = 0.983; RMSEp = 0.937; RPD = 7.566). Taken together, these results demonstrate that a 12-wavelength NIR approach, coupled with optimized preprocessing and machine learning regression, provides a practical and highly accurate alternative for glucomannan assessment, with strong potential for broad implementation—particularly in Indonesia’s porang industry. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2026.