KeyQ Blog

Using machine learning to identify toxic ingredients in products

Written by Eric Nam | Jul 18, 2020 8:17:00 PM

There are over 700 banned chemical ingredients, and over 900 severely restricted ingredients in the European Union.  These chemicals unfortunately still make it into commercial beauty and health products sold to consumers.  Although we may be able to identify a handful of toxic ingredients, and guess the rest by their exotic names, it is a difficult feat for the average daily shopper to look at the ingredients list of a particular product and know exactly what is bad or may cause irritations.

The concern for “unnatural” ingredients is especially important for parents with newborns. Websites, such as EWG’s Skin Deep, help consumers identify restricted or banned chemicals, as well as find information on their toxicity or irritability.  Such tools are important for all shopper who may not have a doctorate degree in chemistry to understand what ingredients are in commercially available beauty and health products.

Eric Nam at KeyQ, who recently became a father, understands the need to identify questionable ingredients before purchasing a product. “As a parent I am constantly concerned about the ingredients we find in products, such as baby lotion.  We need an easy way to catch ingredients that could be harmful and be able to obtain that information readily in a meaningful way,” stated Eric.

Eric leads the development of a new mobile application that uses the camera found on smart phones to “read” the ingredient list on products and identify components that are listed as either banned or severely restricted by the EU. Eric explained “the app is still a prototype in its early stages but we have some very promising progress. Our hope is to refine the user interface to provide an intuitive experience for everyone to quickly identify chemical components from an ingredient list.” He further added that “this technology can be expanded beyond beauty products and used for identifying harmful food additives.”

The application uses machine learning and image vision technology to extract words and identify them through a database. “What makes this unique is you can scan and identify ingredients from any product without relying on the barcode so you are guaranteed a response even if a product doesn’t have a barcode,” explained Eric. The chemical component database will be updated online and downloaded to the app for offline and real-time use.  “We’re working to curate a robust database on known chemicals and ingredients and use machine learning to help create a toxicity score.  Our goal is to make is very easy to understand the potential effects of a particular chemical ingredient,” added Eric.