A machine learning-based clinical decision support system developed using the nine item avoidant/restrictive food intake disorder screen (NIAS)

Avoidant/Restrictive Food Intake Disorder (ARFID) is an eating disorder that can cause significant health issues, particularly in children, yet is often underdiagnosed due to a lack of practical screening tools. The Nine Item Avoidant/Restrictive Food Intake Disorder Screen (NIAS) is a validated and reliable measure for evaluating ARFID symptoms. This study aimed to develop and evaluate a machine learning-based Clinical Decision Support System (CDSS) prototype that classifies ARFID risk based on NIAS-Parent Report data. This study retrospectively analyzed NIAS-Parent Report data collected from 440 children aged 6–12 years in Türkiye. Based on recommended clinical cut-off scores for NIAS subscales, participants were categorized into “High ARFID Risk” or “Low ARFID Risk” groups. Seven tree-based machine learning algorithms were tested using fivefold cross-validation. A Decision Tree model was optimized to prioritize interpretability and clinical feasibility. Among the initial models, Extra Trees and CAT Boosting demonstrated the highest performance (accuracy = 96%). Following optimization, a final Decision Tree model was selected based on its interpretability and simplicity, achieving an overall classification accuracy of 94.1%. The model required 1 to 4 decision points (typically 2–3) for classification. Feature importance analysis revealed NIAS7 (fear-based food avoidance), NIAS4 (low intake/lack of appetite), and NIAS2 (selective eating) as the most predictive items for ARFID risk. The findings suggest that machine learning models, particularly the optimized Decision Tree, demonstrate considerable potential as a CDSS tool for screening ARFID risk in children using NIAS data. This prototype may improve the efficiency and practicality of clinical screening processes. Nevertheless, further studies are needed to establish its external validity and real-world clinical applicability. Avoidant/Restrictive Food Intake Disorder (ARFID) is an eating difficulty where children avoid certain foods or do not eat enough, which can affect their growth, health, and daily life. Early recognition is important, but many healthcare settings still lack quick and practical tools to identify children who may be at risk. The Nine Item ARFID Screen (NIAS) is a short questionnaire that helps parents describe their child’s eating habits, such as strong dislikes of certain tastes or textures, low appetite, or fear of choking or vomiting. In this study, we used NIAS data from more than 400 children aged 6–12 in Türkiye to create a computer-based tool that automatically predicts ARFID risk. We tested several machine learning methods—computer programs that learn from data—and compared how accurately they classified children as “high risk” or “low risk.” We selected a simple and easy-to-understand Decision Tree model so that clinicians can follow the reasoning behind each decision. The final model correctly predicted ARFID risk in about 94% of cases. Questions related to fear-based avoidance, low appetite, and selective eating were the strongest predictors. Overall, this study shows that a machine learning–based tool using NIAS results may help clinicians screen for ARFID more quickly and consistently. Further research is needed to confirm its usefulness in real clinical settings.