Discovery of suicide risk-associated predictors among emerging adults: an interpretable machine learning and network analysis approach

Abstract

Background: Suicide among emerging adults is an important public health concern in China. Identifying key predictors of suicide and examining the associations among these predictors is therefore essential. This study applies a biopsychosocial framework to investigate suicide risk factors in emerging adults. Methods: We adopted a two-wave, six-month longitudinal design and collected questionnaire data via cluster sampling from 681 emerging adults (266 males; M±SD age = 23.98 ± 2.01 years) at a university in Guangdong Province. An XGBoost model combined with Shapley value analysis was used to rank feature importance and select the top 11 predictors. Network analysis was then conducted based on these predictors to explore their interactions. Results: The XGBoost model identified the top 11 predictors of suicide risk. Network analysis revealed that psychological resilience had the highest strength among these predictors, followed by depression, anxiety, and hopelessness. Conclusions: Strengthening psychological resilience, while simultaneously addressing proximal emotional states like depression, anxiety and hopelessness, may reduce the co-occurrence of risk factors. Clinical trial number: Not applicable.