Development of a predictive model for managing lifestyle behaviors among patients with chronic skin diseases: Using machine learning techniques

Abstract

This study aimed to develop a predictive model using lifestyle behavioral factors related to chronic skin disease symptoms and machine learning techniques. A cross-sectional survey was conducted among patients with chronic skin diseases currently receiving treatment at 19 Saengki Korean Medical Clinics specializing in the treatment of chronic skin diseases. Data were collected from 264 participants through an online survey conducted over a three-month period. We used changes in patients' skin symptoms as the dependent variable and lifestyle, behavioral, and treatment variables that affect chronic skin disease remission as predictors. Based on previous studies, we evaluated the performance of the six models using machine learning techniques (decision tree, logistic regression [LR], random forest [RF], CatBoost, gradient boosting classifier, and LightGBM) that are commonly used to create predictive models using categorical factors. The results showed that RF and LR performed well. We selected LR as the final model based on the confusion matrix and receiver operating characteristic (ROC) curve. The LR results showed that herbal medicine use and hospital visits were associated with chronic skin disease symptoms, whereas the RF results showed that herbal medicine use, exercise, and wheat flour consumption were associated with chronic skin disease symptoms. These findings suggest that both the treatment and lifestyle behaviors are associated with chronic skin disease symptoms.