Decoding the Multi-factorial Drivers of Suicidal Motivation in University Students via Interpretable Machine Learning
Year: 2026
Type: article
Study design: Observational Study
Abstract: Suicide among university students remains a major public health concern, yet existing prediction approaches often lack both theoretical depth and practical interpretability. This study aimed to integrate structural equation modeling (SEM) and interpretable machine learning to examine suicidal tendency among Chinese university students within the framework of the Integrated Motivational-Volitional (IMV) model. A total of 3,785 students completed psychological and behavioral measures relevant to suicidal risk. SEM was used to examine theoretically specified associations, and machine learning models were evaluated under three feature configurations. SHapley Additive exPlanations (SHAP) were used to interpret predictor importance. In the primary SEM, defeat was associated with suicidal tendency indirectly through external entrapment, whereas the hypothesized direct association and indirect association through internal entrapment were not supported. Alternative models combining the two entrapment dimensions showed a significant indirect association through overall entrapment. Across eight machine learning algorithms, the comprehensive, intermediate, and simplified configurations achieved mean AUCs of.8570,.8401, and.8453, respectively. A supplementary intermediate configuration including suicide attempt history achieved a mean AUC of.8539, indicating that predictor composition contributed to the comparison. These findings highlight the value of combining theory-driven psychological modeling with interpretable machine learning and support further evaluation of parsimonious predictor sets in independent university samples.
DOI: 10.1145/3862616
Authors: Ran Wu, Jiahao Qin, Jieyu Chen, Yuanfang Wang, Feng Liu
Institutions: East China Normal University, Shanghai Jiao Tong University
Language: English
Subfield: Clinical Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: hybrid
Award: G8791591342
OpenAlex ID: W7225907549
