Background <p>Mental health concerns among college students have received growing attention in China, where routine psychological screening has become part of campus-based mental-health services. However, analyses of these data typically rely on total scores or isolated symptom elevations, paying limited attention to the interrelationships among symptom dimensions.</p> Methods <p>This study applied a Bayesian network framework with simulation-based interventions to examine directional relations among SCL-90 dimensions and to evaluate overall network organization. Data were drawn from the 2024 wave of campus-wide psychological screening (<i>n</i> = 5,979; 54.7% female; <i>M</i>_<i>age</i> = 19.5, <i>SD</i> = 1.8). Nine SCL-90 dimensions were modeled as continuous nodes in a Bayesian network estimated via hill-climbing with 1,000 bootstrap replications; edges were retained using an inclusion threshold ≥ 0.85, with directions determined by majority consensus. Parent nodes identified by out-degree were tested in Gaussian Graphical Model (GGM) and Bayesian network do-simulations. Two complementary outcomes were computed: changes in global strength (ΔGS, structural coupling) and total symptom load (ΔTL, functional burden).</p> Results <p>The averaged Bayesian network contained nine nodes and 26 directed edges. Depression (DEP) and interpersonal sensitivity (INT) emerged as key parent nodes. Attenuating INT yielded the largest ΔGS (–13.47%), reflecting reduced structural coupling, whereas alleviating DEP produced the greatest ΔTL (–45.99%), indicating overall network functional improvement.</p> Conclusions <p>These findings suggest that targeting interpersonal sensitivity may enhance network resilience, whereas achieving overall improvement in psychological functioning requires focusing on depression as the primary intervention pathway. The present study offers a data-driven framework for guiding future prevention and intervention efforts in college mental-health settings.</p>