Investigating the role of mental health and lifestyle behaviours in the risk of adolescent suicidal behaviours through machine learning
- 1Shanghai Civil Aviation College, Shanghai, China.
- 2Shanghai SIPO Polytechnic, Shanghai, China.
Abstract
Background: Suicidal behaviors among adolescents are a serious global public health concern and are closely linked to both mental health and everyday lifestyle behaviors. Suicidal ideation, planning, and attempts lie along a spectrum of increasing severity. However, how psychological distress and health-related behaviors interact to influence suicide risk is complex and not yet fully understood. This study aimed to identify key mental health and lifestyle factors associated with different stages of adolescent suicidal behaviors using interpretable machine learning methods. Methods: Data were obtained from the 2021 Youth Risk Behavior Surveillance System (YRBS), a nationally representative survey of U.S. high school students. After data cleaning and multiple imputation, 16,946 adolescents aged 14–18 years were included in the analysis. Extreme Gradient Boosting (XGBoost) models were developed to predict suicidal ideation, suicide planning, and suicide attempts. Model performance was assessed using the area under the receiver operating characteristic curve (AUC) and related metrics. Feature importance was examined using gain-based measures, while SHapley Additive exPlanations (SHAP) and partial dependence plots were used to explore the direction, strength, and nonlinear effects of key predictors. Results: All three XGBoost models showed satisfactory predictive performance. Mental health factors, particularly persistent feelings of sadness and self-rated mental health, were the strongest predictors across all stages of suicidal behaviors. Lifestyle behaviors, including sleep duration, physical activity, diet, screen time, and substance use, also contributed meaningfully to risk prediction, although their relative importance differed by outcome. SHAP analyses indicated that greater sadness and poorer mental health were strongly associated with higher suicide risk, whereas adequate sleep and higher levels of physical activity were associated with lower predicted risk. Substance use, including alcohol, cigarette, and marijuana use, played an important role in predicting suicide attempts than suicidal ideation or planning. Partial dependence analyses showed nonlinear and threshold effects, especially for sadness, mental health status, and sleep duration. Conclusion: Mental health factors, especially persistent sadness and overall psychological well-being, were among the most prominent factors associated with suicidal behaviors among adolescents, while lifestyle behaviors showed important and stage-specific associations. Interpretable machine learning approaches help identify complex and nonlinear patterns associated with adolescent suicide risk and offer valuable insights to guide targeted, evidence-based prevention efforts.
Related
- Cultural adaptation and validation of the Suicidal Behaviours Questionnaire – Autism Spectrum Conditions (SBQ-ASC) for use in Brazil
- Effect of nursing care program on healthy lifestyle behaviours and quality of life in women with endometriosis: a randomised controlled trial
- Self-rated health and lifestyle behaviours differences in second-generation migrants and Italians: an analysis of the daily life aspects surveys in Italy, 2015–2022
- Brain functional changes of suicidal ideation and behaviours: a meta-analysis of task-based fMRI studies.
- Recent Clinical Encounters for Suicidal Thoughts and Behaviours: Disclosure, Assessment, Missed Opportunities, and Implications for Practice
BibTeX
@article{ding2026investigating,
title = {{Investigating the role of mental health and lifestyle behaviours in the risk of adolescent suicidal behaviours through machine learning}},
author = {Yanmei Ding and Yanzhao Lei},
journal = {PLOS One},
year = {2026},
month = oct,
doi = {10.1371/journal.pone.0358758},
url = {https://doi.org/10.1371/journal.pone.0358758},
note = {CC BY},
}
RIS
TY - JOUR
TI - Investigating the role of mental health and lifestyle behaviours in the risk of adolescent suicidal behaviours through machine learning
AU - Yanmei Ding
AU - Yanzhao Lei
PY - 2026
DA - 2026/10/21
JO - PLOS One
DO - 10.1371/journal.pone.0358758
UR - https://doi.org/10.1371/journal.pone.0358758
AB - Background: Suicidal behaviors among adolescents are a serious global public health concern and are closely linked to both mental health and everyday lifestyle behaviors. Suicidal ideation, planning, and attempts lie along a spectrum of increasing severity. However, how psychological distress and health-related behaviors interact to influence suicide risk is complex and not yet fully understood. This study aimed to identify key mental health and lifestyle factors associated with different stages of adolescent suicidal behaviors using interpretable machine learning methods. Methods: Data were obtained from the 2021 Youth Risk Behavior Surveillance System (YRBS), a nationally representative survey of U.S. high school students. After data cleaning and multiple imputation, 16,946 adolescents aged 14–18 years were included in the analysis. Extreme Gradient Boosting (XGBoost) models were developed to predict suicidal ideation, suicide planning, and suicide attempts. Model performance was assessed using the area under the receiver operating characteristic curve (AUC) and related metrics. Feature importance was examined using gain-based measures, while SHapley Additive exPlanations (SHAP) and partial dependence plots were used to explore the direction, strength, and nonlinear effects of key predictors. Results: All three XGBoost models showed satisfactory predictive performance. Mental health factors, particularly persistent feelings of sadness and self-rated mental health, were the strongest predictors across all stages of suicidal behaviors. Lifestyle behaviors, including sleep duration, physical activity, diet, screen time, and substance use, also contributed meaningfully to risk prediction, although their relative importance differed by outcome. SHAP analyses indicated that greater sadness and poorer mental health were strongly associated with higher suicide risk, whereas adequate sleep and higher levels of physical activity were associated with lower predicted risk. Substance use, including alcohol, cigarette, and marijuana use, played an important role in predicting suicide attempts than suicidal ideation or planning. Partial dependence analyses showed nonlinear and threshold effects, especially for sadness, mental health status, and sleep duration. Conclusion: Mental health factors, especially persistent sadness and overall psychological well-being, were among the most prominent factors associated with suicidal behaviors among adolescents, while lifestyle behaviors showed important and stage-specific associations. Interpretable machine learning approaches help identify complex and nonlinear patterns associated with adolescent suicide risk and offer valuable insights to guide targeted, evidence-based prevention efforts.
SN - 1932-6203
ER -