Constructing a Bayesian neural network model for the cross-sectional classification of psychological distress among older adults: comprehensive analysis of humor styles, communication skills, and physical activity experiences.

Objectives: This study aimed to develop a model for the cross-sectional classification of psychological distress among older adults, using a Bayesian Neural Network (BNN) to examine the relationships among humor expressions, communication skills, physical activity (PA), and social PA experiences (with friends), and their effects on psychological distress risk. Methods: A cross-sectional survey was conducted among 5,265 Japanese adults aged 65 to 89 years. The predictor variables included humor expressions, communication skills, and social PA experiences. Depression was assessed using the K6 Psychological Distress Scale (cutoff ≥ 9). A BNN with three hidden layers was constructed with SHapley Additive exPlanations (SHAP) for feature importance identification. Results: The BNN model achieved 81.1% accuracy for high-risk detection. This study revealed self-enhancing humor coping and self-control communication skills as the strongest protective factors. Additionally, past PA experiences with friends and present PA experiences alone reduced the risk for K6-assessed psychological distress. Conclusion: The BNN model identified positive humor styles, self-controlled communication abilities, past social PA, and present PA status as important predictors of contributors to the model output for classifying K6-assessed psychological distress among older adults, highlighting the potential relevance of social engagement in communication and PA.