Development of a Short-Form Hwa-Byung Symptom Scale Using Machine Learning Approaches

Abstract

Background/Objectives: Hwa-byung (HB), also known as “anger syndrome” or “fire illness”, is a culture-bound syndrome primarily observed among Koreans. This study aims to develop a short-form version of the HB symptom scale using machine learning approaches. Methods: Utilizing exploratory factor analysis (EFA) and various machine learning techniques (i.e., XGBoost, Logistic Regression, Random Forest, Support Vector Machine, Decision Tree, and Multi-Layer Perceptron), we sought to create an efficient HB assessment tool. A survey was conducted on 500 Korean adults using the original 15-item HB symptom scale. Results: The EFA revealed two distinct factors: psychological symptoms and somatic manifestations of HB. Statistical testing showed no significant differences between using different numbers of items per factor (ANOVA: F = 0.8593, p = 0.5051), supporting a minimalist approach with one item per factor. The resulting two-item short-form scale (Q3 and Q10) demonstrated high predictive power for the presence of HB. Multiple machine learning models achieved a consistent accuracy (90.00% for most models) with high discriminative ability (AUC = 0.9436–0.9579), with the Multi-Layer Perceptron showing the highest performance (AUC = 0.9579). The models showed balanced performance in identifying both HB and non-HB cases, with precision and recall values consistently around 0.90. Conclusions: The findings of this study highlighted the effectiveness of integrating EFA and artificial intelligence via machine learning in developing practical assessment tools. This study contributes to advancing methodological approaches for scale development and offers a model for creating efficient assessments of Korean medicine.