Chemometric and meta-heuristic algorithms to find optimal wavelengths and predict ‘Red Delicious’ apples traits using Vis-NIR

• Vis-NIR spectra collected by optic probe at two orthogonal points on the apple. • Linear and non-linear models’ development for quality parameter prediction. • Effective wavelength selection based on meta-heuristic algorithms. • Optimised prediction of apple quality parameters: from ascorbic acid to firmness. The potential of Vis-NIR spectroscopy (540–960 nm) in quality evaluation of Red Delicious apples was investigated. This approach can contribute to the development of portable quality assessment devices to estimate quality parameters including pH, TA, ascorbic acid, firmness, soluble solids content (SSC), and anthocyanins using machine learning. Good performances were obtained for all the tested regressors (PLSR, PCR, MLR, SVM-R, and ANN) in terms of R 2 , RMSE, and RPD values in cross-validation, validation, and test phases. SVM-R demonstrated slightly higher performance in predicting apples traits in validation: pH (R 2 0.935, RMSEP 0.019, and RPD 3.636), TA (R 2 0.932, RMSEP 0.026, and RPD 3.447), SSC (R 2 0.913, RMSEP 0.174, and RPD 3.168), ascorbic acid (R 2 0.938, RMSEP 0.154, and RPD 3.725), firmness (R 2 0.932, RMSEP 0.311, and RPD 3.522), and anthocyanin (R 2 0.921, RMSEP 0.007, and RPD 3.171). Thus, different metaheuristic variable selection techniques (including ACO, LA, GA, PSO, LCA, FOA, WCC, CUK, DSOS, ICA, and HTS) were combined with SVM to identify the most important wavelengths for practical industry applications. SVM-PSO and SVM-FOA were identified as the most effective wavelength selection methods based on average coefficient of correlation, average convergence of error, execution time, and the number of selected wavelengths.