소아 허약 검사를 활용한 허약아 패턴 분석 및 머신러닝 기반 예측 모델 개발

Abstract

Objectives: This study aimed to explore the internal structure of the Pediatric Weakness Scale (PWS) through factor analysis, classify pediatric weakness patterns using hierarchical clustering, and evaluate a machine-learning-based predictive model for clinical screening. Methods: PWS responses from 326 children aged 6–9 years without organic diseases were analyzed. Factor analysis (principal axis factoring and varimax rotation) was used to extract seven latent factors. These factor scores were used for hierarchical cluster analysis (Ward’s linkage method and Euclidean distance), resulting in ten clusters each. Cluster characteristics were compared using Multidimensional Fatigue Scale (MFS), Korean Child Behavior Checklist (K-CBCL), Korean version of Children’s Eating Behavior Questionnaire (K-CBEQ), respiratory infection history, and growth indices. Machine learning models were trained to predict clusters based on factor scores, and their performance was evaluated using 10-fold cross-validation. Results: Seven factors were identified: respiratory health, general condition, ocular health, emotional regulation, gastrointestinal function, nutrition, and sleep-related nervous system sensitivity. Ten distinct clusters were derived: healthy children; severe, moderate, and mild complex weakness; liver–lung weakness; emotional dysregulation – heart weakness; neurosensitive heart weakness; and Spleen, Lung, and Kidney – specific weakness. Among the classifiers, the nonlinear support vector machine achieved the highest accuracy (0.890). Conclusion: This study suggests a new classification and prediction model for pediatric weakness based on PWS data. These results support the potential for early screening and personalized health strategies in pediatric care.