Development of Prediction Model for Suicide Attempts Using the Korean Youth Health Behavior Web-Based Survey in Korean Middle and High School Students
10.4306/jknpa.2023.62.3.95
- Author:
Younggeun KIM
1
;
Sung-Il WOO
;
Sang Woo HAHN
;
Yeon Jung LEE
;
Minjae KIM
;
Hyeonseo JIN
;
Jiyeon KIM
;
Jaeuk HWANG
Author Information
1. Department of Psychiatry, Soonchunhyang University Hospital Seoul, Seoul, Korea
- Publication Type:ORIGINAL ARTICLE
- From:Journal of Korean Neuropsychiatric Association
2023;62(3):95-101
- CountryRepublic of Korea
- Language:English
-
Abstract:
Objectives:Assessing the risks of youth suicide in educational and clinical settings is crucial.Therefore, this study developed a machine learning model to predict suicide attempts using the Korean Youth Risk Behavior Web-based Survey (KYRBWS).
Methods:KYRBWS is conducted annually on Korean middle and high school students to assess their health-related behaviors. The KYRBWS data for 2021, which showed 1206 adolescents reporting suicide attempts out of 54848, was split into the training (n=43878) and test (n=10970) datasets. Thirty-nine features were selected from the KYRBWS questionnaire. The balanced accuracy of the model was employed as a metric to select the best model. Independent validations were conducted with the test dataset of 2021 KYRBWS (n=10970) and the external dataset of 2020 KYRBWS (n=54948). The clinical implication of the prediction by the selected model was measured for sensitivity, specificity, true prediction rate (TPR), and false prediction rate (FPR).
Results:Balanced bag of histogram gradient boosting model has shown the best performance (balanced accuracy=0.803). This model shows 76.23% sensitivity, 83.08% specificity, 10.03% TPR, and 99.30% FPR for the test dataset as well as 77.25% sensitivity, 84.62% specificity, 9.31% TPR, and 99.45% FPR for the external dataset, respectively.
Conclusion:These results suggest that a specific machine learning model can predict suicide attempts among adolescents with high accuracy.
