Unraveling Social Network Factors in Predicting Depression with a Machine Learning Approach

This study identifies the key factor contributing to major depressive disorder using a machine learning approach. Depression is a global public health concern, particularly significant in South Korea due to its strong association with high suicide rates. While demographic, socioeconomic, medical history and social network-focused factors are associated with depression, the consensus on the most critical one is challenging due to methodological limitations. To address this, we applied Partial Least Squares Discriminant Analysis (PLS-DA) and evaluated selectivity ratios. 172 participants were included, 70 depressed and 102 non-depressed, assessed by the Hamilton Depression Rating Scale. To gauge the social embeddings of participants, we used UCLA Loneliness Scale (UCLA-3). We included demographic, socioeconomic, and medical history features for the all-inclusive model. We found that the social network related factors were more critical than others. Seven items from the UCLA, including “No one really knows me well”, had a selectivity ratio greater than 2. No features from other factors were found significant. This study underscores that poor-quality social relationships are strongly associated with depression. These findings can enhance early screening for depression and enable the development of tailored interventions for effective treatment and management.