Answers are grounded in this paper and cite their sources.
Owing to insufficient information in the Standard Table of Food Composition in Japan (STFCJ), estimation of sugar intake has proven difficult in Japan. Using machine learning models, we aimed to predict total sugar content in foods with unlisted sugars in the STFCJ. We used data on 10 nutrients from the STFCJ or the U.S. Food and Nutrient Database for Dietary Studies (FNDDS). The top 10 nutrients with the highest contribution rates, identified based on SHAP values, were selected as features. The STFCJ dataset was clustered based on the 10 nutrients used for training, followed by data augmentation, and models were developed for each cluster. We compared the regression version of five machine learning models: support vector regression (SVR), random forest (RF), extra trees (ET), and multilayer perceptron regression (MLP). The ET model exhibited the highest R 2 among the models trained on the STFCJ dataset (0.850) and the FNDDS dataset (0.573). Of all models, the ET model trained on the clustered STFCJ dataset exhibited the highest R 2 value (0.983). The models trained on the STFCJ dataset exhibited higher R 2 values than those trained on the FNDDS dataset. These findings show that machine learning models can accurately predict total sugar content, providing a novel approach to estimate sugar intake in Japan.
No takes yet. Share an insight, caveat, or question.
Torami et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: