It has been reported repeatedly that depression in middle-aged people may cause serious ramifications in public health. However, previous studies on this important research topic have focused on utilizing either traditional statistical methods (i.e., logistic regressions) or black-or-gray artificial intelligence (AI) methods (i.e., neural network, Support Vector Machine (SVM), ensemble). Previous studies lack suggesting more decision-maker-friendly methods, which need to produce clear interpretable results with information on cause and effect. For the sake of improving the quality of decisions of healthcare decision-makers, public health issues require identification of cause and effect information for any type of strategic healthcare initiative. In this sense, this paper proposes a novel approach to identify the main causes of depression in middle-aged people in Korea. The proposed method is the Sons and Spouses Bayesian network model, which is an extended version of conventional TAN (Tree-Augmented Naive Bayesian Network). The target dataset is a longitudinal dataset employed from the Korea National Health and Nutrition Examination Survey (KNHANES) database with a sample size of 8580. After developing the proposed Sons and Spouses Bayesian network model, we found thirteen main causes leading to depression. Then, genetic optimization was executed to reveal the most probable cause of depression in middle-aged people that would provide practical implications to field practitioners. Therefore, our proposed method can help healthcare decision-makers comprehend changes in depression status by employing what-if queries towards a target individual.
Keywords: Bayesian network; KNHANES; depression; genetic algorithm; health informatics; machine learning.