Predicting Depressive Disorders in Diabetic Workers: A Comparative Analysis of Relevance Vector Machine and Traditional Machine Learning Models

Abstract

This study aimed to identify key variables associated with depressive disorders in middle-aged workers with diabetes (n=609) using the Relevance Vector Machine (RVM) and to compare the performance of various machine learning models, including RVM, CART, SVM, and C-SVM. Analysis of variable importance using RVM revealed that perceived stress, poor self-rated health status, gender (female), age (40-49), and educational attainment (high school graduate or below) were significant factors. Notably, perceived stress and poor self-rated health status had the highest importance, indicating their substantial impact on depressive disorders. The RVM model showed superior performance across most metrics, achieving the highest ROC AUC of 0.78, signifying high classification performance in predicting depressive disorders.Future research should include diverse data and analyze variable interactions to address these limitations. The results provide foundational data for mental health management in diabetic workers, emphasizing the need for tailored intervention strategies and effective approaches for depressive disorders prevention and management.