Food Data Analysis using Multidimensional Visualizations based on Point Placement

Abstract

Food data comprise records regarding nutrients, ingredients, amounts of different vitamins and minerals that can be found in foods. The wide variety of food products that can be stored in large datasets makes the traditional analysis tasks unfeasible and time-consuming when conducted manually by the dietitians and related professionals. This paper describes a method for visualizing food data using point placement strategies to support specialists in tasks related to determining similar food products that can be replaced in specific diets. The proposed method generates a structured representation for food data to be used as input to some state-of-the-art and recent visualizations, such as PCA, t-SNE, UMAP and TriMap. Experiments were conducted to assess the quality of visualizations and the results reported that the nonlinear visualizations presented satisfactory discriminability regarding some food categories and better preservation of the data patterns. A case study based on a visual exploration process was also conducted and demonstrates the specialist successfully finding substitute food products for planning a vegan diet plan.