Detecting and Preventing Food Fraud Using Data Analytics
Whether it’s fake olive oil, coffee bulked up with husks and twigs, or honey tainted with antibiotics, food fraud is a growing problem worldwide. The Australian research organization CSIRO states that the economic damage alone from food fraud has reached $35 billion (in US dollars) in 2018. The underlying cause is nearly always financial gain and economic pressures to save money by using inferior (or mislabeled) products. Predictive analytics is one tool manufacturers are using to combat food fraud.
Even in countries with strict labeling regulations, a number of food manufacturers or suppliers are deceiving distributors and the public by adding fillers, diluting the real product with less expensive ingredients or misrepresenting their products.
This article is posted on our Science Snippets Blog.
Is it real or fake? Multivariate statistical tools are useful in analyzing data with numerous sources of variance (such as NIR spectroscopy) and other add new methods of detecting food fraud.
See examples using data analytics
It’s become so prevalent that the U.S. Pharmacopeial Convention, a nonprofit organization that sets standards used by the FDA, set up a Food Fraud Database (FFD). Now operated by Decernis, the Food Fraud Database describes food fraud as the: “deliberate substitution, addition, tampering or misrepresentation of food, food ingredients or food packaging, or false or misleading statements made about a product for economic gain.”
More recently, the CEN (European Committee for Standardization) working group goes straight to the point and defines food fraud as: “to intentionally cause a mismatch between food product claims and characteristics.”
The FFD has a shocking number of entries. Some of the most common foods subjected to food fraud are olive oil, milk, honey, saffron, orange juice, coffee, apple juice, grape wine, vanilla extract, and maple syrup. For consumers who cook with olive oil, the food safety threat is tremendous, as peanut or hazelnut oil, both common allergens, are often substituted. But the problems are not just health-related.
Food safety issues aside, supplier and distributor mislabeling creates economic and reputation threats as well. One study by Oceana based on seafood samples from 674 retail outlets across the U.S. from 2010 to 2012 found that 33 percent of the 1,215 samples they analyzed were mislabeled according to FDA guidelines. Most commonly highly desirable or expensive seafood products had been replaced with less desirable or cheaper ones.
Generally, the two main types of economically-motivated food fraud are:
- Selling food that is inferior and potentially harmful, such as: recycling animal by-products back into the food; packing and selling meat with unknown origins; knowingly selling goods past their ‘use by’ date.
- Deliberately mislabeling food, such as: substituting products with a cheaper alternative (such as farmed salmon sold as wild, or Basmati rice adulterated with cheaper varieties) or making false statements about the source of ingredients (for example, geographic, plant or animal origin).
Shaun Kennedy of the U.S. National Center for Food Protection and Defense says, “About 10 percent of the food you buy in the grocery shelf is probably adulterated."
Image source: Food Safety Magazine.
Regulatory bodies such as the FDA and FSIS will soon make “food defense plans” required by food processors under their jurisdiction. This ranges from regulations to prevent bioterrorism, health risk and patent or labeling fraud.
See examples using data analytics
Using Science to Detect Food Fraud
Methods used to detect counterfeit products often involve “chemical fingerprinting” or “gene detection.” These methods include using technology such as:
- Near-infrared spectroscopy
- X-ray spectroscopy
- Mass spectrometry
- Chromatography, liquid and gas
- Wet chemistry, like atomic absorption spectroscopy (AAS) or one of the inductively coupled plasma (ICP) variants
- DNA testing
Similar to how the ridgelines of the skin in fingerprints can be measured to provide a unique identity, chemical fingerprinting uses the geochemistry of the environment to determine the geographic origin of a product. This is done by comparing light-stable isotopes (carbon, nitrogen, sulphur, oxygen, hydrogen) and/or trace elements in food product ingredients to the density of the same elements at the claimed origin. DNA testing is e.g. used to analyze meat or fish proteins to determine the species used in the food product.
By measuring the absorption of near-infrared light in a substance, near-infrared spectroscopy provides a physicochemical fingerprint of a biological sample. The fingerprint may contain up to 1,000 spectral variables that each relate to the physicochemical composition of the food in their own unique way.
In addition, NIR spectroscopy allows large quantities of raw materials or ingredients to be measured at once, making it a viable technique to be used during the manufacturing process. With spectroscopic monitoring it is possible to examine nearly 100% of the ingredients and raw materials that go into production, thereby considerably reducing production errors or outputs that are of a lower quality than the formulation allows.
“The problem is that the food analyses which are predominantly used today are only spot checks and they are typically targeted towards a single kind of food fraud. We would like to move away from this old-school methodology and instead take a non-targeted physicochemical fingerprint of the foodstuffs. By using fingerprints and contrasts we can determine whether a given batch of raw materials or ingredients are defective or different compared to the usual,” said Professor Søren Balling Engelsen, the head of the section Chemometrics & Analytical Technology within the Food Science Department (FOOD) at Copenhagen University, Denmark in an article.
By using chemical fingerprints and contrasts, combined with multivariate data analytics, food manufacturers can determine whether a given ingredient is what it claims to be.
What Can Manufacturers Do?
The first step may be assessing risk. How vulnerable are you to food fraud? Factors such as the ingredient market price, its fraud history, events affecting availability (e.g. failed crops through draughts, flooding, storms, etc.), composition, physical state and level of processing determine how inherently vulnerable a particular food line might be.
As a rule of the thumb, more highly processed ingredients, for example, apple juices or apple purees may be more vulnerable than raw materials like apple pieces. Assessing vulnerability shouldn’t be a one-time activity, however, but something done on a continuous or frequent basis.
After assessing risk, manufacturers can establish a plan for mitigation measures, which may include establishing specifications for their desirable raw materials (including the chemical fingerprints or as an alternative UV absorption rates), creating a monitoring plan (both of factory processes and raw materials), auditing suppliers and establishing an alert system when problems are detected.
Analytics methods for monitoring raw material to verify authenticity may include:
- Targeted analyses (linked to parameters specified in raw material specifications)
- Untargeted techniques (finger-printing) that assess the raw material integrity against adulteration
Predictive Analysis Helps Prevent Food Fraud
Although many companies have huge amounts of “big data” available about their manufacturing processes, they aren’t always using it effectively, or preventively. In a 2016 article in Food Manufacture, Rick Pendrous said, “Brand owners need to use predictive analysis of business data to help prevent them being victims of food fraud.”
At a conference on “The Future of the Food Industry” in February 2016, Jude Mason, director for consulting and technical services at NSF International, said her company uses predictive analytics to focus on areas of risk and “inform the future.”
One way companies can do this is through the use of “chemometrics.” This type of analytics provides statistical and mathematical approaches that can extract useful information from large and complex datasets such as chemical data and reactions. Multivariate statistical tools are useful in analyzing this type of data, which is comprised of values and properties of various compounds with numerous sources of variance (such as NIR spectroscopy).
In Comprehensive Reviews In Food Science And Food Safety Free Access, D. Granato, et al., explain how they use pattern recognition methods, such as principal component analysis and cluster analysis, can be used to associate the level of bioactive components with in vitro functional properties, and how multivariate statistical methods can be used for authentication purposes across a broad range of food fraud authentication applications.
Some of the statistical approaches used to help classify products and determine whether food fraud exists, are:
- discriminant analysis (DA), which is similar to logistic regression, and
- one-class classifiers (OCC), or class modeling methods.
DA methods can confirm whether or not an object can be associated with a targeted class of interest. For example, using this method, it’s possible to answer the questions: “does this olive oil come from Italy or Spain?” or “is this milk from organic or conventional sources?”
Another method that doesn’t use any type of distribution assumption and can be used for small samples, is the nearest k-NN or k-nearest neighbor method that can be applied to classification of categorical data variables and regression for continuous variables. With this method, it’s possible, for example to determine that two honey samples were produced in different geographical origins.
The OCC methods are intensively applied in food chemistry for quality control and authentication of various foods, for example, for classification of olive oils, quality control of peanut oils and fruit juices.
Using Data Analytics Tools to Distinguish Types
SIMCA®, a multivariate data analytics tool can be used to help bring clarity to multifactor data and discover food fraud. They have other uses in the food and beverage industry as well.
Find Out More
Want to know more about how data analytics can be used to improve food and beverage manufacturing?
Download this paper highlighting four case examples.