基于SVM的出行方式特征分析和识别研究

Travel Mode Character Analysis and Recognition Based on SVM

基于智能手机采集的居民出行轨迹信息,分析了不同出行方式的特征,利用支持向量机进行了出行方式识别研究.首先探讨了利用手机软件所能检测和记录的参数,进而从出行轨迹和特征参数两个方面对出行方式特征进行了分析,探讨了不同出行方式两两可分的关键变量,提取用于识别不同出行方式的特征向量,最后建立了径向基核函数支持向量机(SVM)分类器.利用从大连市出行轨迹数据获取的出行方式样本,训练了该支持向量机,并且以决策树、BP神经网络为对照.结果表明,SVM识别精确度为89.6%,BP神经网络为85.5%,决策树为77.3%,SVM具有更好的识别性能.

This paper focuses on travel mode recognition based on Support Vector Machines（SVM） after analyzing the characters of different travel mode, which are extracted from travel trace information collected by smartphone. Firstly, we figure out which parameters can be detected and recorded by smartphone, and analyze the character of travel mode from two aspects which are travel trace and character parameters, to find which are the key parameters to divide each two kind of travel mode, so that we are able to make up the character vectors to recognize different travel mode, and eventually a classifier of Radial Basis Function based SVM can be established. This SVM is trained by travel mode samples from travel trace data in Dalian. As comparisons, Decision Tree and BP Neural Network are used. The result shows that the recognition accuracy rate of SVM is 89.6%, which of BP Neural Network is 85.5%, which of Decision Tree is 77.3%, and suggests that the SVM has better recognition performance.