Predicting Adolescent Suicidal Ideation: The Role of Physical Activity Evaluated through Machine Learning Models

Background and Purpose

In the U.S., suicide is the second leading cause of death among adolescents. Adolescence is a risk-prone period, marked by impulsivity and emotional volatility, which can fuel suicidal thoughts. Although physical activity may mitigate this risk, its impact in the context of cyberbullying and increased screen time—a challenge of the digital age—warrants further investigation. In this study, we used XGBoost, a machine learning method, to analyze Youth Risk Behavior Survey (YRBS) data and predict the effects of risk factors on youth suicidal ideation. Physical activity emerged as a key variable, consistently registering the highest F-score and demonstrating significant predictive importance across models. This work indicates that machine learning can enhance suicide prevention strategies by underscoring the protective role of physical activity for young people in today's digital landscape.

Research Questions:

How do cyberbullying and screen time impact suicidal ideation in adolescents?

Could consistent physical activity offset the effects of cyberbullying and screen time on adolescents' suicidal thoughts?

Methods

This study utilized the 2021 YRBS to examine health risk behaviors among U.S. high school students in grades 9 through 12. To predict suicidal ideation, the study analyzed data from 3,760 students, focusing on demographic variables such as age, gender, sexual orientation, and race/ethnicity. Additionally, it considered factors related to youths' physical activity, screen time, cyberbullying, depression, and being overweight to predict suicidal thoughts in adolescents. The variables were explored using the XGBoost algorithm running in Google Colab and statistical analysis in SPSS 27.0.

Results

The XGBoost model correctly predicted 3,008 cases and inaccurately predicted 752 cases, achieving an accuracy rate of 75.93%. The XGBoost model's analysis indicates that depression is a primary node, significantly influencing initial predictions of suicidal ideation. Leaf node values suggest a strong correlation between an increased likelihood of suicidal thoughts and experiences of depression, cyberbullying, or minority sexual identity, with the highest risk level reflected by a value of 0.426880735. In contrast, the negative value of -0.0577540025 intimates a just-below threshold risk. Physical activity (PHY_ACT_7), with the top F score of approximately 345 in the YRBS dataset, emerges as a critical predictor, indicating its overarching influence in the model. Target plots correlate minimal physical activity with the highest ideation risk (0.410) and the most active group with the lowest (0.207), accentuating the potential of physical activity as a significant factor in mental health interventions for adolescents at risk of suicide.

Conclusions and Implications

This study highlights the important role of physical activity in reducing suicidal thoughts among youth and urges stakeholders such as educational institutions, healthcare providers, community organizations, and families to integrate physical activity into mental health strategies. It advocates for policies and programs that prioritize physical activity to prevent youth suicide. In accordance with the goal of ‘Strengthening Social Impact through Collaborative Research,' this study shows the value of collaborations between social scientists and data scientists to understand the factors predicting suicide in adolescents, thereby contributing to the Grand Challenge of Social Work, ‘Ensuring Healthy Development for Adolescents.’