BackgroundHidden hunger (HH) refers to the deficiency of certain micronutrients. Current research suggests that approximately 70% of chronic diseases are linked to HH, which significantly affects public health. Consequently, there is an urgent need for an effective method to assess the risk of HH. This study aims to develop risk prediction models for HH using machine learning (ML).MethodsWe conducted a questionnaire survey among 9336 high school students in 11 cities within Anhui Province and assessed their HH risk using a scale. After quality control, we designated 632 students from Xuancheng City as the external test cohort and used the remaining 6477 students as the training cohort to develop predictive models. We used six ML algorithms (i.e., deep‐learning neural network [DNN], random forest, support vector machine, extreme gradient boosting, gradient boosting decision tree, and k‐nearest neighbor) to fit the training set using five‐fold cross‐validation, with hyperparameter tuning performed via Bayesian optimization. We used the “Streamlit” library to construct an online application and the “shapley additive explanations” library for model interpretability analysis.ResultsWe observed that the DNN model performed best. In the external test cohort, the area under the curve reached 0.813, accuracy was 0.739, and sensitivity and specificity were 0.720 and 0.760, respectively. Furthermore, the precision‐recall curve, calibration curve, and decision curve analysis also indicated that our model had high predictive accuracy. To aid practical use, we developed an online application (http://sec.mitusml.com:9000/). Through model interpretability analysis, we discovered that the frequent consumption of fruits and coarse grains was likely to reduce the risk of HH, whereas frequently eating snacks and fried foods increased the risk of HH.ConclusionsWe developed an effective prediction model for HH and analyzed the factors that influence its risk.
