基于机器学习简化MMPI量表的有效性研究  

Research on the Validity of Simplifying MMPI Scale Based on Machine Learning

作　　者：孙启科 董问天[3] 王克[4] 冯超南 崔霖 李毓明 于淏岿 于滨 石川[3] 纪俊 SUN Qi-ke;DONG Wen-tian;WANG Ke;FENG Chao-nan;CUI Lin;LI Yu-ming;YU Hao-kui;YU Bin;SHI Chuan;JI Jun(College of Computer Science&Technology,Qingdao University,Qingdao 266071,China;Medical College,Qingdao University,Qingdao 266071,China;Peking University Sixth Hospital,Beijing 100083,China;Qingdao Municipal Hospital,Qingdao 266011,China;Beijing Wanling Pangu Technology Co.,Ltd.,Beijing 100089,China)

机构地区：[1]青岛大学计算机科学技术学院,青岛266071 [2]青岛大学医学部,青岛266071 [3]北京大学第六医院,北京100083 [4]青岛市市立医院,青岛266011 [5]北京万灵盘古科技有限公司,北京100089

出　　处：《青岛大学学报（自然科学版）》2021年第2期23-28,共6页Journal of Qingdao University(Natural Science Edition)

基　　金：国家自然科学基金(批准号:61503208)资助;山东省自然科学基金(批准号:ZR2015PF002)资助。

摘　　要：针对传统MMPI量表中题目数量过多,许多应用场景下受试者依从性差的问题,提出使用机器学习算法对MMPI量表中的临床量表进行简化的方法,对6种经典机器学习算法的简化效果进行了比较,并在受试者测评数据上对原始和简化后测评结果的敏感度与特异度进行了对比分析。实验结果表明,使用最小绝对收缩和选择算子(LASSO)的算法在保持相近准确率的同时简化效果最好,平均减少了37.3%的男性和39.1%的女性受试者题目数量,并且简化后测评结果的敏感度与特异度仍能保持在原始测评结果的85%,缩短了测评时间,适用于更多的应用场景。There are too many questions in the traditional MMPI scale and the subjects’compliance is poor in many application scenarios.A machine learning algorithm is proposed to simplify the clinical scale in the MMPI scale.The simplification effects of six classical machine learning algorithms are compared,and the sensitivity and specificity of the original and simplified evaluation results are compared and analyzed on the subject evaluation data.The experimental results show that the algorithm using the minimum absolute contraction and selection operator(LASSO)has the best simplification effect while keeping similar accuracy to reduce the number of subjects by 37.3%for men and 39.1%for women on average,keep the sensitivity and specificity of the simplified evaluation results at 85%of the original evaluation results,thus shortening the evaluation time and being suitable for more application scenarios.

关 键 词：MMPI 机器学习 量表简化 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]