There has been a growing concern in utilizing machine learning models to identify risk factors for adolescent mental health. However, the comorbidity domain has not received adequate attention. Accordingly, this study aims to develop an efficient machine leaning model to predict the comorbidity of depression and self-injury among adolescents. 1,028,751 Chinese adolescents completed measures of depression, self-injury, and a range of items related to sociodemographic and psychosocial variables. We evaluated the performance of six machine learning models and established the optimal model for identifying the comorbidity of depression and self-injury. We selected the Top-N variable set corresponding to a cumulative probability of 80% for the optimal model to establish a risk model for the comorbidity of depression and self-injury in adolescents. The combined model of Random Forest and LightGBM can effectively identify adolescents with comorbidity risk based on 13 variables. Specifically, the predictive power of individual characteristics significantly outweighs environmental factors
within individual characteristics, emotional problems (anxiety) exhibit the strongest predictive power
among environmental factors, parental emotional maltreatment and cyber victimization demonstrate the highest predictive effect. This study extends the application of the Bioecological Model in the field of comorbidity research, demonstrating the advantages of using machine learning methods to predict comorbidity of depression and self-injury in adolescents. It holds practical value for preventing and intervening in comorbidity of depression and self-injury among adolescents.