基于人格特质和机器学习分类算法的建筑工人不安全行为识别  

Identification of Unsafe Behaviors of Construction Workers Based on Personality Traits and Machine Learning Classification Algorithms

作　　者：周建亮 胡飞翔 邢艳冬 高嘉瑞 袁华凯 ZHOU Jian-liang;HU Fei-xiang;XING Yan-dong;GAO Jia-rui;YUAN Hua-kai(School of Mechanics and Civil Engineering,China University of Mining and Technology,Xuzhou 221116,China;Xuzhou Design Institute,China Railway Shanghai Design Institute Group Corporation,Xuzhou 221116,China)

机构地区：[1]中国矿业大学力学与土木工程学院,徐州221116 [2]中铁上海设计院集团有限公司徐州设计院,徐州221116

出　　处：《科学技术与工程》2022年第29期13013-13020,共8页Science Technology and Engineering

基　　金：国家自然科学基金(72171224);教育部人文社科规划基金(19YJAZH122);江苏省研究生科研与实践创新计划(KYCX22_2680);中国建设教育协会教育教学科研课题重点项目(2021168)。

摘　　要：获取建筑工人个性特征是实现其不安全行为精准化、个性化干预管理的重要前提,而人格特质是分析建筑工人个性特征的重要依据。以292名一线建筑工人为研究对象,通过问卷调研和深度访谈探究人格特质与不安全行为之间的映射关系,基于大五人格生成不安全行为偏好,利用机器学习分类算法实现建筑工人的不安全行为识别。研究表明:高外倾性、中神经质、中宜人性、低责任心、低开放性映射习惯偏差型不安全行为;中外倾性、低神经质、低宜人性、低责任心、高开放性映射程序偏差型不安全行为;中外倾性、高神经质、中宜人性、高责任心、中开放性映射感知偏差型不安全行为;中外倾性、中神经质、中宜人性、中责任心、中开放性映射技能偏差型不安全行为。同时通过比选分类与回归树(classification and regression tree, CART)、随机森林(random forest, RF)、自适应提升树(adaptive boosting, AdaBoost)和梯度提升决策树(gradient boosting decision tree, GBDT)四种分类算法模型的评估指标,结果发现GBDT算法的不安全行为预测性能最优。Obtaining construction workers’ personality characteristics is an important prerequisite for achieving precise and personalized intervention management of their unsafe behaviors, and personality traits are an important basis for analyzing construction workers’ personality characteristics. 292 front-line construction workers were selected as the research objects. The mapping relationship between personality traits and unsafe behavior was explored by questionnaire survey and in-depth interview. Based on the big five personality, unsafe behavior preferences were generated, and machine learning classification algorithm was used to identify unsafe behavior of construction workers. The results show that high extraversion, moderate neuroticism, moderate agreeableness, low responsibility, and low openness mapping habit-biased unsafe behaviors. Moderate extraversion, low neuroticism, low agreeableness, low responsibility, and high openness mapping procedural-biased unsafe behaviors. Moderate extraversion, high neuroticism, moderate agreeableness, high responsibility, and moderate openness mapping perceptual-biased unsafe behaviors. And moderate extraversion, moderate neuroticism, moderate agreeableness, moderate responsibility, and moderate openness mapping skill-biased unsafe behaviors. Meanwhile, by comparing the evaluation metrics of four classification algorithm models, classification and regression tree(CART), random forest(RF), adaptive boosting(AdaBoost) and gradient boosting decision tree(GBDT), it is found that the GBDT algorithm has the best performance for unsafe behavior prediction.

关 键 词：建筑工人 不安全行为 人格特质 分类算法 行为识别 

分 类 号：TU714[建筑科学—建筑技术科学]