An automated non-destructive prediction of peroxide value and free fatty acid level in mixed nut samples

Abstract

This study aimed to develop an automated technique, which is rapid, non-destructive and inexpensive, to test for rancidity of nuts. A visible to near infrared benchtop hyperspectral camera was used to capture images from blanched canarium, unblanched canarium and macadamia samples. Support vector machine classification (SVC) and PLSR models were developed to segregate the pooled spectra of the nuts and predict their peroxide values (PV) and free fatty acid (FFA) concentrations. The SVC and PLSR models were then used in a hierarchical model to develop an automated system for predicting PV and FFA. The automated model was then tested using a test set providing classification accuracy of 87% and R2 between 0.60 and 0.76 and RPD between 1.6 and 2.7 for PV and FFA prediction. Overall, the automated system has the potential commercial application in nut processing to detect rancidity of mixed nut samples non-destructively and in real-time. It is suggested to train other machine learning models with more samples to improve the accuracy of predictions.