Detecting fraudulent additions in skimmed milk powder using a portable, hyphenated, optical multi-sensor approach in combination with one-class classification

Judith Müller-Maatsch,Martin Alewijn,Michiel Wijtten,Yannick Weesepoel

doi:10.1016/j.foodcont.2020.107744

Abstract

The detection of fraudulent additions to milk powder is an ongoing research subject for governmental agencies, industry and academia. Current developments steer towards the application of so-called fingerprint approaches, describing authentic, reference samples with spectroscopy and using one-class classification (OCC) to identify “out-of-class”, or adulterated samples. Within this article we describe the application of a novel, portable device hyphenating ultraviolet–visible, fluorescence and near-infrared spectroscopy in combination with OCC modelling to discriminate authentic skimmed milk powders from adulterated ones. As adulterated samples we analyzed skimmed milk powder with the addition of plant protein powder, whey powder, starch, lactose, glucose, fructose as well as non-protein nitrogen like ammonium chloride, ammonium nitrate, melamine and urea in different concentrations. After fusion of the classification results from the three spectral techniques and several models two scenarios are presented. 100% (scenario 1) or 80% (scenario 2) of the authentic skimmed milk powders were correctly identified as “in-class”, whereas respectively 64% or 86% of the adulterated samples were correctly classified as “out-of-class”. In brief, this article provides insights in the application of novel, portable devices that may be applied in a non-invasive manner and gives an outlook on data handling and a new data fusion strategy.

Highlights

In food authenticity and food safety testing, targeted analysis of hazardous compounds is increasingly replaced with fingerprinting analysis (Gao et al, 2019; Riedl, Esslinger, & Fauhl-Hassek, 2015)
We report the usage of a novel, portable, hy phenated sensor in the detection of food fraud that generates informa tion from different spectroscopic approaches at the same time from the same sample and spot
The multivariate statistics approach used enables a tailored application of different thresholds to balance the false negative or false positive classifications targeting the needs of the respective operators such as governmental agencies, industry and academia

Summary

Introduction

In food authenticity and food safety testing, targeted analysis of hazardous compounds is increasingly replaced with fingerprinting analysis (Gao et al, 2019; Riedl, Esslinger, & Fauhl-Hassek, 2015). In addition to the shift of analytical methodology in food fraud detection, measurements are preferred to be performed on-site and in a non-invasive and fast manner. This drives the development of portable devices that carry miniaturized optical spectrometers (Croc ombe, 2018; Ellis, Muhamadali, Haughey, Elliott, & Goodacre, 2015; McGrath et al, 2018; Yeong, Jern, Yao, Hannan, & Hoon, 2019). The combination of the data, i.e. fusion of spectra or statistical output is believed to give a more accurate fingerprint of the sample (Callao & Ruisanchez, 2018). For fast spectroscopic applications in food sensing, data needs to

Methods

Results

Conclusion