Identifying Non-Linear Predictors of Suicidal Ideation in Hospitalized Youth using SHAP-explained Extreme Gradient Boosting
Journal of Adolescent and Youth Psychological Studies · 2026 · 7 (3) · 1-11
DOI: 10.61838/kman.jayps.5152 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Adolescent and Youth Psychological Studies, 2026. Verfasst von Petrus Amutenya und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Objective
The objective of this study is to identify and interpret the non-linear clinical and demographic predictors of suicidal ideation among hospitalized youth using a SHAP-explained Extreme Gradient Boosting predictive model. Methods and Materials: A retrospective cross-sectional study was conducted using the electronic medical records of hospitalized youths aged to years in Nigeria. Clinical and demographic data, including evaluations from the Columbia-Suicide Severity Rating Scale, Patient Health Questionnaire (PHQ-9), Generalized Anxiety Disorder scale (GAD-7), and Adverse Childhood Experiences (ACE) count, were extracted. An Extreme Gradient Boosting (XGBoost) machine learning algorithm was trained to predict the presence of suicidal ideation. To resolve the “black-box” nature of the algorithm, the SHapley Additive exPlanations (SHAP) framework was utilized to interpret model outputs, rank global feature importance, and map complex, non-linear risk thresholds via dependence plots.
Findings
The overall prevalence of suicidal ideation in the clinical cohort was ( ). The XGBoost predictive model demonstrated excellent diagnostic performance on the independent test set, achieving an accuracy of and an Area Under the Curve (AUC) of . SHAP analysis identified depressive severity as the most dominant global predictor, revealing a distinct non-linear threshold where predictive risk exponentially spiked once a PHQ-9 score reached exactly . The cumulative burden of trauma emerged as the second strongest predictor, with or more adverse childhood experiences significantly amplifying risk, specifically interacting with older adolescence (ages to ). Furthermore, anxiety demonstrated a bimodal, non-monotonic risk curve, while a history of non-suicidal self-injury within the past months formed a steep logarithmic risk trajectory.
Conclusion
Traditional linear clinical assessments fail to capture the complex, exponential tipping points of youth suicide risk. Integrating SHAP-explained XGBoost algorithms into psychiatric practice can uncover critical non-linear thresholds.
Zitieren
APA
Amutenya, P., Alvarado, S., & Vaultier, M. (2026). Identifying Non-Linear Predictors of Suicidal Ideation in Hospitalized Youth using SHAP-explained Extreme Gradient Boosting. Journal of Adolescent and Youth Psychological Studies, 7(3), 1-11. https://doi.org/10.61838/kman.jayps.5152
Vancouver
Amutenya P, Alvarado S, Vaultier M. Identifying Non-Linear Predictors of Suicidal Ideation in Hospitalized Youth using SHAP-explained Extreme Gradient Boosting. Journal of Adolescent and Youth Psychological Studies. 2026;7(3):1-11. doi:10.61838/kman.jayps.5152
BibTeX
@article{Amutenya2026,
author = {Amutenya, Petrus and Alvarado, Sofía and Vaultier, Manuel},
title = {Identifying Non-Linear Predictors of Suicidal Ideation in Hospitalized Youth using SHAP-explained Extreme Gradient Boosting},
journal = {Journal of Adolescent and Youth Psychological Studies},
year = {2026},
volume = {7},
number = {3},
pages = {1--11},
doi = {10.61838/kman.jayps.5152}
}
EUVIMED