This page was translated automatically from Turkish. Read the original

This page is an AI-assisted Turkish summary of the study presented at the 26th World Congress of Psychiatry. The study belongs to the presenting authors; please refer to the original source when citing it.

EV1310

An explainable machine learning approach to predicting suicidal ideation in university students: The SMS-ME project

AN EXPLAINABLE MACHINE LEARNING APPROACH TO PREDICT SUICIDAL THOUGHTS IN UNIVERSITY STUDENTS. THE SMS-ME PROJECT.

An explainable XGBoost model predicted suicidal ideation among healthcare students in internal and external samples with AUROC values ranging from 0.72 to 0.81; the strongest predictors were depression, stress, anxiety, and episodes of vomiting.

The aim was to develop a clinically interpretable and generalizable explainable machine learning (ML) model to identify suicidal ideation among university students in Sicily by integrating academic, psychological, and behavioral predictors. An anonymous online survey was conducted between September 2022 and June 2024 among students enrolled in medicine, nursing, psychology, dentistry, and other healthcare programs at universities in Sicily; sociodemographic and academic information, DASS-21 and EAT-26 scores, and suicidal ideation over the preceding six months were recorded. An XGBoost model was used to predict suicidal ideation, with medical and nursing students forming the training set and students in other programs forming the external validation set; missing data were imputed using KNN, class imbalance was addressed using BorderlineSMOTE, performance was assessed through nested cross-validation using AUC-ROC, accuracy, sensitivity, specificity, and F1 score, and SHAP (SHapley Additive exPlanations) analysis was applied to support interpretability. A total of 3,278 healthcare students were included (69.5% female; mean age 22.7 ± 4.3 years), and 26.2% reported suicidal ideation or a suicide attempt in the preceding six months. Severe or extremely severe symptoms were reported by 34.8% for depression, 52.6% for anxiety, and 37.1% for stress, while 14.2% of students were at risk of an eating disorder. The XGBoost model demonstrated good predictive performance in internal and external samples, with AUROC values ranging from 0.72 to 0.81; SHAP analysis identified depression, stress, anxiety, and episodes of vomiting as the most influential predictors. In conclusion, explainable artificial intelligence may complement clinical decision-making by providing transparent and interpretable risk assessment for early identification and targeted prevention in university settings; university-specific predictors related to academic performance and eating behavior highlight the influence of academic pressure and body image concerns.

All WCP 2026 summaries