Comparative Performance of Machine Learning Models Using Food Intake Frequency Versus Vegetable Intake Data to Predict Problematic Mealtime Behaviour in Japanese Preschool Children
Abstract
Background
Selective eating often leads to nutritional imbalance and mealtime stress, making early identification essential.
Objective
This study aimed to compare the predictive performance of machine learning (ML) models using two types of input data—namely, vegetable intake and food frequency questionnaire (FFQ) responses—to identify children with selective eating behaviour (cutoff score ≥1.59).
Material and methods
We analysed a cross-sectional dataset of 283 children aged 3–6 years using up to 8 predictors selected from 15 food-frequency items or 15 designated vegetables, along with age and sex. Selective eating behaviour was defined as a selective eating score of ≥ 1.59. Eleven ML algorithms were trained and evaluated, resulting in 11,253 feature–model combinations. Model performance was assessed using 5-fold cross-validation based on accuracy, precision, recall, F1 score, and ROC-AUC. Permutation importance analysis was conducted to identify key predictive features.
Results
Among the 11 ML algorithms tested, vegetable intake-based L1 logistic regression showed the best performance (ROC-AUC = 0.717, accuracy = 0.647, precision = 0.722, recall = 0.659, F1 = 0.808). Permutation importance further identified tomato, green onion, and taro as key contributors in the vegetable intake-based L1 logistic regression model. In contrast, the FFQ-based Naïve Bayes model showed relatively good discrimination (ROC-AUC = 0.619) despite moderate recall.
Conclusion
Vegetable intake-based ML models were more effective than those using FFQ data in identifying children with selective eating behaviour. Detailed dietary assessment aids early detection.
© 2026 Naoki Sakane, Yaeko Kawaguchi, Junichiro Somei, Akiko Suganuma, Masayuki Domichi, published by Institute of Mother and Child
This work is licensed under the Creative Commons Attribution 4.0 License.