KMAN Publication Inc. (KMANPUB) Journal of Personality and Psychosomatic Research (JPPR) 3041-8542 4 Serial Number 15 2026 07 01 Multimodal Machine-Learning Prediction of Eating-Related Psychosomatic Risk Based on Impulsivity, Interoceptive Deficits, Emotional Eating, and Body Mass Index Multimodal Machine-Learning Prediction of Eating-Related Psychosomatic Risk Based on Impulsivity, Interoceptive Deficits, Emotional Eating, and Body Mass Index 1 21 EN Journal Article 2026 04 12 <p>This study aimed to develop, compare, and interpret multimodal machine-learning models for predicting elevated eating-related psychosomatic risk among Mexican adults based on impulsivity, interoceptive deficits, emotional eating, and body mass index. A cross-sectional predictive study was conducted with 812 adults recruited from five metropolitan areas in Mexico. Impulsivity was assessed using the Barratt Impulsiveness Scale–11, interoceptive deficits using the corresponding subscale of the Eating Disorder Inventory–3, emotional eating using the Dutch Eating Behavior Questionnaire, and body mass index through standardized height and weight measurements. Elevated eating-related psychosomatic risk was defined as the simultaneous presence of clinically relevant disordered eating attitudes and moderate or greater somatic symptom burden. The sample was stratified into a development set of 650 participants and an independent test set of 162 participants. Logistic regression, elastic-net regression, support vector machine, random forest, extreme gradient boosting, and early- and late-fusion multimodal neural networks were evaluated using nested cross-validation. Emotional eating, interoceptive deficits, impulsivity, and body mass index were significantly and positively associated with the continuous psychosomatic-risk index. The early-fusion multimodal neural network produced the strongest test-set performance, with an ROC–AUC of .905, precision–recall AUC of .742, sensitivity of .882, specificity of .875, balanced accuracy of .879, and Brier score of .098. It significantly outperformed conventional logistic regression. Ablation analyses showed that excluding emotional eating produced the largest reduction in predictive accuracy, followed by removing interoceptive deficits and impulsivity. Body mass index alone showed limited discrimination, whereas its addition modestly improved the psychometric model. Eating-related psychosomatic risk was predicted most accurately through integration of emotional, interoceptive, behavioral, and anthropometric information, with emotional eating and interoceptive deficits representing the most influential predictors.</p> https://www.journals.kmanpub.com/index.php/jppr/article/download/5944/11368