Evaluation of machine learning algorithms -k nearest neighbors and support vector machines- for strawberries classification during food drying process

Abstract

In the present work, supervised machine learning (ML) algorithms, k-NN and SVM, were applied to classify strawberry samples during a microwave-assisted drying process. A dataset of 1150 strawberry records containinginformation about images of two pre-treatments types (fresh, FR and osmotically pre-treated, OD), three ranges of drying times (short <40 min; intermediate: 40-70 min and long> 70 min) and three physical characteristics previously selected (shrinkage, brightness and saturation) was used to perform the ML classifiers. The k-NN and SVM models led to good accuracy values, 0.94 for sample type and 0.90 for drying time categories. Since colour and morphological changes are related to changes in the product quality, these results are useful to evaluate the losses of nutritional and sensorialproperties taking place during the in-line processing of strawberries, by means of a non-invasive monitoring technique.