Feasibility of chemometric and machine learning strategy for lignin type identification using FT-IR spectroscopic data

Abstract Proto-lignin and technical lignin differ in functional groups, degree of condensation, and purity. Rapid and reliable classification between these types supports optimized valorization in biorefineries. This study presents a chemometric and machine learning approach for the reliable classification of lignin types using Fourier-transform infrared (FT-IR) spectroscopic data. FT-IR spectra, in combination with advanced preprocessing and classification algorithms, were employed to distinguish between proto-lignin and technical lignin with high accuracy (>83 %). Among the tested classifiers, partial least squares discriminant analysis (PLS-DA) demonstrated superior performance in terms of accuracy, precision, sensitivity, and specificity. Extensive model optimization was performed using various spectral pretreatment techniques, and the combination of standard normal variate (SNV) with Savitzky–Golay (S–G) first-derivative filtering yielded the highest classification efficiency. FT-IR spectral data covering the full range (4,000–650 cm −1 ) and its three segmented regions were evaluated separately; however, the full-range spectra produced the most robust model, achieving R 2 values of 99.35 % for calibration and 93.30 % for validation. The proposed methodology offers a rapid, non-destructive, and cost-effective screening tool for distinguishing lignin types, providing a valuable analytical strategy for lignin characterization and quality control in biorefinery applications.