青少年心理危机随机森林预测模型的建立及影响因素分析

The establishment of a random forest predictive model and analysis of influencing factors for psychological crisis among adolescent

目的基于机器学习随机森林算法建立青少年心理危机预测模型,分析青少年心理危机的影响因素。方法分别在2020年11月与2021年6月,采用整群抽样追踪调查1417名中学生,第一次测量收集人口学资料、症状因素、保护因素等问卷数据,第二次测量抑郁、自杀风险,以是否在第二次测量中呈现中度以上抑郁(抑郁得分≥15分)与高自杀风险(自杀风险得分≥7分)为心理危机判定标准。运用SPSS 24.0进行统计学分析,采用R version 4.1.1软件构建青少年心理危机随机森林机器学习预测模型,并分析青少年出现心理危机的高预估因素。结果(1)中度以上抑郁检出率为10.02%(142/1417),高自杀风险检出率为30.77%(436/1417),心理危机检出率为8.19%(116/1417)。(2)心理危机预测模型敏感度为0.79,特异度为0.82,阳性预测值为0.82,阴性预测值为0.79,准确率为0.80,曲线下面积为0.88。(3)青少年心理危机影响因素排名前十的特征变量依次为抑郁情绪、焦虑情绪、自杀意念、自我伤害行为、认知灵活性-可控性、认知灵活性-可选择性、坚毅-坚持努力、坚毅-兴趣一致性、母亲情绪和父亲情绪(模型预测精准度=0.023~0.163)。结论青少年心理危机的发生与症状因素、保护因素、父母情绪关系密切,且有跨时间预估的意义。机器学习随机森林算法能有效识别心理危机个体,识别敏感的危机个体特征。

Objective To establish a predictive model of psychological crisis based on the machine learning random forest algorithm,and to analyze the influencing factors of psychological crisis among adole scent.Methods A total of 1417 middle school students were surveyed using cluster sampling in two phases,in November 2020 and June 2021.Demographic data,symptom factors,protective factors were collected in the first investigation,and depression and suicide risk were measured in the second investigation.The criteria for psychological crisis were moderate to severe depression(depression score≥15)and high suicide risk(suicide risk score≥7)in the second measurement.SPSS 24.0 software was used for statistical analysis of variables,and the random forest machine learning predictive model for psychological crisis was established by using R version 4.1.1 software,and the high-estimating factors of adolescent psychological crisis were analyzed.Results (1)The detection rate of moderate to severe depression was 10.02%(142/1417),the detection rate of high suicide risk was 30.77%(436/1417),and detection rate of the psychological crisis was 8.19%(116/1417).(2)The sensitivity and specificity of psychological crisis prediction model were 0.79,0.82,positive predictive value was 0.82,negative predictive value was 0.79,accuracy was 0.80 and area under curve was 0.88.(3)The top 10 characteristic variables of influencing factors of adolescent psychological crisis were depression,anxiety,suicidal ideation,self-harming behavior,cognitive flexibility-controllability,cognitive flexibility-selectivity,grit-persistence effort,grit-interest consistency,mother's mood and father's mood(model prediction accuracy was 0.023-0.163).Conclusion sThe occurrence of adolescent psychological crisis is closely related to symptom factors,protective factors and parental emotions,and has the significance of predicting across time.The machine learning random forest algorithm can effectively identify psychological crisis individuals and identify sensitive crisis individual characteristics.