A machine learning-based predictive model of nonsuicidal self-injurious behavior among college students in Guizhou Province
10.16835/j.cnki.1000-9817.2023.08.018
- VernacularTitle:基于机器学习构建贵州省大学生非自杀性自伤行为的预测模型
- Author:
PAN Chan, LIU Xiaorong, SHI Xiangzi, ZHAO Wenxin, TIAN Meng, CHEN Siyuan, ZHANG Wanzhu
1
Author Information
1. School of Public Health, Guizhou Medical University, Guiyang (550025) , China
- Publication Type:Journal Article
- Keywords:
Selfinjurious behavior;
Mental health;
Forecasting;
Students
- From:
Chinese Journal of School Health
2023;44(8):1198-1202
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the effectiveness of machine learning algorithms in predicting non-suicidal self-injury (NSSI) behavior among college students, and to analyze the influencing factors of NSSI behavior, thus providing a reference for promoting psychological well-being.
Methods:In December 2022, a stratified random cluster sampling method was used to select 835 college students from a university in Guizhou Province, China. The Adolescent Self-injury Scale, Family Function Assessment Scale, and Emotion Regulation Self-efficacy Scale were used to evaluate the participants. Demographic characteristics, family factors, and emotional factors were taken as independent variables, while the dependent variable was whether college students exhibited NSSI behavior. Machine learning algorithms, including Logistic regression, support vector machine (SVM), decision trees, algorithm gradient boosting trees, random forests, and AdaBoost, were used to construct predictive models.
Results:The detection rate of NSSI behavior among the college students was 23.23% (194 individuals). The NSSI behavior group scored higher than the non-NSSI behavior group in total family function, emotional communication, egoism, and family rules ( t=3.02, 3.35 , 2.23,2.87, P <0.05). On the other hand, the non-NSSI behavior group scored higher than the NSSI behavior group in total emotion regulation selfefficacy, managing negative emotion self-efficacy, and expressing positive emotion self-efficacy ( t=-5.04, -5.48 , -2.43, P <0.05). The recall rates of random forests, SVM, Logistic regression, decision trees, algorithm gradient boosting trees, and AdaBoost were 84.3% , 90.6%, 73.4%, 87.5%, 95.3%, 89.0%, respectively. The F1 scores were 84.4%, 92.1%, 71.2 %, 79.4%, 91.7%, 89.1% , respectively. The respective precision rates were 84.4%, 93.5%, 69.1%, 72.7%, 88.4%, 89.1 %. The AUC scores were 0.845, 0.922, 0.706, 0.776, 0.915, and 0.891, respectively.
Conclusion:Compared to the algorithm gradient boosting tree, random forest, Logistic regression, and AdaBoost models, the SVM model has a better predictive effect on whether college students in Guizhou Province exhibits NSSI behavior. It is recommended to use an appropriate model to identify students at risk of NSSI behavior as early as possible and provide psychological crisis interventions to promote their mental health.