Automated identification of healthier food substitutions through a combination of graph neural networks and nutri-scores

While maintaining a healthy diet can be challenging in the current food environment, food substitutions can offer individuals the opportunity to replace less nutritious foods with products having a more balanced nutrient composition. To this end, we have developed an approach that can recommend healthier alternative food products in individuals' current diets. The first part of the algorithm automatically generates food substitutions that are in the same food category than the query food using Graph Neural Networks (e.g., machine learning model designed to make predictions on graphs), while the second part ranks these food options according to their highest Nutri-Score. The specific results achieved were a Recall Rate of 0.647 and 0.733 for the top 5 and top 10 results, respectively; meaning that our algorithm finds more than 65% of the relevant ingredients in the top-5 recommended food substitutions and 75% of the relevant ingredients in the top-10 recommended food sub-stitutions, showing promising outcomes. However, one major drawback of our approach is that it is less effective in discovering food substitutions that are not closely related as our algorithm restricts the recommended food item to be in the same food category than the query food.