Establishment of a Near-Infrared Spectroscopy-Based Screening Framework for Key Quality Indicators of Cassava Varieties

Cassava is the sixth most important food crop globally, valued for its high starch accumulation in tuberous roots, which supports diverse applications in food processing and industrial production. Quality evaluation is vital for its utilization, yet rapid and efficient assessment methods remain underexplored. In this study, 67 cassava varieties were collected to assess their nutritional quality. Near-infrared spectroscopy (NIRS) in the wavenumber range of 11,000–4000 cm−1 was applied to predict key quality traits of cassava flour from different varieties. Combined with partial least squares (PLS), principal component analysis (PCA), and internal cross-validation, a preliminary NIRS-based screening framework for cassava flour quality indicators was established. The protein and moisture models achieved an excellent quantitative prediction performance (Rcv2 > 0.82, RPD > 4.0), qualifying them as reliable tools for routine analysis. In contrast, the models for starch, amylose, and amylopectin showed substantial overfitting with low cross-validation accuracies (Rcv2 = 0.19–0.38), restricting their use to only preliminary screening. The fat model performed poorly (Rcv2 = 0.16, RPD = 1.08), rendering it unsuitable for any quantitative or screening application. This study provides an exploratory screening framework for rapid multi-index evaluation of cassava flour quality. However, the marked variability in predictive performance across constituents highlights the critical need for external validation to improve model robustness.