Machine learning‐based risk warning for adolescents' prosocial behavior

Prosocial behavior plays a positive role in promoting adolescents' mental health and behavioral performance. Guided in ecological systems theory and social cognitive theory, this study used machine learning to identify the risk factors of adolescents' prosocial behavior, and a total of six machine learning algorithms (Logistic Regression, Naive Bayes, Decision Tree, Random Forest, KNN, and LightGBM) were tested and compared to detect the 55 potentially risk factors (e.g., individual, community, family, school, and social) of prosocial behavior. The sample includes 8,364 adolescents (age varied from 13 to 15; 47.9% girls) from middle and high schools in Zhejiang province, China, collected by a multi-stage cluster random sampling. The logistic regression algorithm and LightGBM algorithm are better than the other four algorithms. By comparing the relative importance of each factor, the results of the study suggest that individual perceived relative deprivation and negative attachment type have a higher value in predicting prosocial behavior, which can help adolescents in developing positive social behavior and provide stronger support for social adjustment. This study contributes to understanding and revealing the influencing factors and mechanisms of adolescents' prosocial behavior among adolescents. Its applied value lies in helping schools and counselors identify adolescents whose low prosocial behavior may reflect difficulties in attachment or internet addiction, thereby informing early support for adolescent social well-being.