Nutrient profile-based food categorization and group-wise missing data imputation for commercial food composition database

Food composition databases (FCDBs) are essential resources for personalized nutrition, product development, and food substitution analysis. However, their applicability is often limited by missing data and heterogeneity among food classification systems. In this study, we propose a data-driven framework for nutrient-based food categorization and missing data imputation. First, we apply clustering algorithms to group food items based on nutritional similarity, providing a complementary approach to conventional food classification systems. This nutrition profile-based food group supports interoperability and reduces the manual effort required for mapping heterogeneous food categories. Second, we introduce a group-wise imputation strategy that incorporates food group information to improve the estimation of missing nutrient values. Using commercial data from two different FCDBs, we show that incorporating nutrient-based food group information into the imputation process yields more reliable results than conventional methods that do not exploit such structure. Furthermore, we evaluate the effectiveness of deep learning-based imputations that capture complex nutrient relationships, even in the absence of explicit group information. The proposed methodology offers a scalable and low-cost solution for improving food category systems and nutrient data coverage, contributing to more coherent and analytically useful FCDBs. • Nutrient-profile based clustering can complement food categories in FCDBs. • Group-wise imputation using food groups improves missing nutrient estimation. • Real branded food data were used to validate the proposed approach. • Clustering-derived labels outperform traditional categories in key tasks. • Deep imputation model achieve high accuracy on nutrient data.