A machine-learning algorithm used to automatically quantify the processing level of any food (artificial alterations of food nutrients due to cooking, packaging, and additives), given its nutrient profile, with the goal to classify all foods in the food supply. Up to now, the assessment of food processing has been done manually, for just 34.25% of the food supply.
Technology Overview
Northeastern researchers trained a multi-class random forest to automatically predict the processing level of any food, given its logarithmic nutrient profile with the goal to classify all foods in the food supply. For each food item, the algorithm provides for probabilities: 
- Likelihood to be unprocessed or minimally processed
- Likelihood to be a processed culinary ingredient
- Likelihood to be processed 
- Likelihood to be ultra-processed
The algorithm assigns each food to the class with the highest probability. As training data used in this invention has been manually labeled subset of the food supply classified by NOVA. Inventors have performed 5-fold cross-validation over the labeled database, obtaining excellent performances in four classes of AUC and AUP.
- Cost-effective/Greater efficiency: the algorithm overcomes manual classification and efficiently scales to big data. 
- Greater reliability: the classification is supported by reliability scores (probabilities), absent in the manual classification. 
- Adaptability: the algorithm can be easily adapted to different nutrient lists describing foods in different databases.
- Automatic assessment of the processing level of any food
- Analysis of the whole food supply and its changes in time for public health assessment
- Analysis of the individual diet to evaluate the dietary intake of processed foods
- Recommendation tool for consumers: comparison of the processing level of the same food developed by different brands
- A recommendation tool to suggest cooking or preserving methodologies that minimally alter raw ingredients 
- Product development: Can systematically test which recipe variations make a food product less processed
- License
- Partnering
- Research collaboration
Patent Information:
For Information, Contact:
Mark Saulich
Associate Director of Commercialization
Northeastern University