基于机器学习的公交卡数据中通勤人群辨识方法  

An Approach to the Identification of Commuters Based on Machine Learning of Smartcard Data

作　　者：孙世超 庄斌 黄伟 

机构地区：[1]北京晶众智慧交通科技股份有限公司,北京100012

出　　处：《交通工程》2017年第1期58-64,共7页Journal of Transportation Engineering

基　　金：国家自然科学基金(51478350):基于大数据关联分析的城市公交建设与改善决策支持研究

摘　　要：以厦门市公交IC卡数据为依托,提出一种基于问卷调查数据的机器学习分类方法——决策树模型,用于分类辨识公交刷卡数据中的通勤人群.研究中利用公交刷卡数据与调查问卷中同时包含的公交出行信息作为非类别属性变量,以公交通勤人群/非公交通勤人群作为分类变量,分别构建决策树模型中的树节点与叶节点.之后,利用随机抽取的2/3调查样本数据构成决策树算法的训练集,余下数据作为决策树算法的测试集,对模型进行构建、训练与测试,结果显示模型预测效果良好,其覆盖率约为98.1%,命中率也达到81%,最终将该模型用于公交刷卡数据中通勤人群的分类辨识.This paper applied the Decision Tree algorithm to identify transit commuters, based on thedata of smartcard and questionnaire survey in Xiamen. Firstly, we chose three attributes as theinterviewer 爷 s non-category variables and one attribute as the category variable （ commuters/noncommuters）,by the use of the bus travel information which were contained in both the questionnaire andsmartcard data. Then the tree was established and trained based on two-thirds of the survey data. Byusing the rest of questionnaire sample that was not used in the training, we then tested the predictionaccuracy of the calibrated model. The cover rate of the prediction was 98. 1 % and the accuracy rate is81%. Finally, the validated decision tree model was applied to identify transit commuters from thesmartcard data, the result shows that the number of transit commuters in Xiamen may be around 300 000.

关 键 词：城市交通 IC卡数据 决策树分类方法 通勤人群 

分 类 号：U491.17[交通运输工程—交通运输规划与管理]