Investigation of pre-processing NIR spectroscopic data and classification algorithms for the fast identification of chocolate-coated peanuts and sultanas

Chocolate-coated confectionery, including fruits and nuts, is an increasingly popular snack food. Non-destructive discrimination of the core composition could be useful for quality assurance purposes, such as ensuring the absence of peanuts in a batch of chocolate-coated sultanas. This study investigated the optimum pre-processing methods and discrimination algorithms for identifying chocolate-coated peanuts and sultanas from their near-infrared (NIR) spectra. The best-performing results were found using partial least squares discriminant analysis (PLS-DA) and principal component analysis with linear discriminant analysis (PCA-LDA), which both demonstrated 100% classification accuracy when applied to the validation set. Principal component analysis with support vector machine (PCA-SVM) showed slightly poorer results, particularly when using non-optimal pre-processing techniques. In general, the most accurate results were found when using either the unprocessed or SNV-processed spectral data. This work supports the prospect of using near-infrared spectroscopy for the quality assurance in the manufacture or wholesale of panned chocolate goods.