机器学习在绘画测验预测青少年依赖型人格偏离中的应用  

Application of machine learning in the prediction of drawing test on dependent personality dysfunction in adolescents

作　　者：陈涛 王鹏翀 林轩 裴欢昌 邢怡伦 罗捷 项锦晶[1] 王亚[2] CHEN Tao;WANG Pengchong;LIN Xuan;PEI Huanchang;XING Yilun;LUO Jie;XIANG Jinjing;WANG Ya(School of Humanities and Social Sciences,Beijing Forestry University,Beijing 100083,China;CAS Key Laboratory of Mental Health,Institute of Psychology,Beijing 100101,China;Beijing Key Laboratory of Mental Disorders,Capital Medical University,Beijing 100088,China)

机构地区：[1]北京林业大学人文学院,北京100083 [2]中国科学院心理健康重点实验室(中国科学院心理研究所),北京100101 [3]首都医科大学附属北京安定医院,精神疾病诊断与治疗北京市重点实验室,北京100088

出　　处：《中国心理卫生杂志》2019年第10期769-773,共5页Chinese Mental Health Journal

基　　金：中央高校基本科研业务费专项资金资助(2019RW13);北京市社会科学基金项目“投射技术对中小学生立体化心理评估模式探索”(15JYC033)

摘　　要：目的:运用机器学习探索绘画测验对青少年依赖型人格偏离的预测力。方法:研究样本是318例青少年,采用人格障碍诊断问卷的依赖型人格障碍分量表来评定青少年的依赖型人格偏离,借助统合型"房树人"绘画测验分析其绘画特征。共筛选出依赖型人格偏离阳性组79例,阴性组239例。运用机器学习探讨11项绘画特征对依赖型人格偏离的预测性能。结果:在对青少年依赖型人格偏离的预测中,所选取的11项绘画特征中,树干、伤痕(0.20)、人靠近树(0.18)和中心画(0.13) 3个特征的平均重要性最高。机器学习模型预测的准确率为0.87,精度为0.85,召回率为0.86,F1分数为0.85。结论:绘画测验结合机器学习能较好地预测青少年依赖型人格偏离。Objective:To explore whether drawing test can effectively predict the dependent personality dysfunction with machine learning.Methods:A total of 318 adolescents were selected from our previous study.The Dependent Personality Diagnostic sub-scale of Personality Diagnostic Questionnaire-4+ was adopted to measure the dependent personality dysfunction in these participants.The Synthetic House-Tree-Person(S-HTP) test was used to analyze the drawing characteristics of the participants.Seventy-nine adolescents were classified into positive group and 239 were classified into negative group.With machine learning,11 drawing characteristics were used to predict the dependent personality dysfunction among teenagers.Results:When predicting dependent personality dysfunction in adolescents with machine learning,scars on the trunk(0.20),the person near the tree(0.18) and the central picture(0.13) were found to show the highest average importance ratio among the 11 drawing features.The model has 87% accuracy,85% precision,86%recall,and 85% F1 score,which indicated an acceptable performance of this machine learning model in predicting dependent personality dysfunction in adolescents.Conclusion:It suggests that together with machine learning,drawing test could classify the dependent personality dysfunction effectively.

关 键 词：机器学习 随机森林算法 人格偏离 绘画测验 青少年 

分 类 号：B844.2[哲学宗教—发展与教育心理学] B848[哲学宗教—心理学]