Research On Urban Residents’ Travel Mode Identification Based On Mobile Phone Signaling Data

With the rise of Big data applications,the analysis of residents’ travel behavior is becoming more and more diversified.Due to the continuous innovation of positioning technology and the widespread popularity of smartphones,mobile signaling data is increasingly becoming an important data source in the study of residents’ travel behavior analysis.Many scholars have conducted research on the extraction of travel trajectories,analysis of travel characteristics,and analysis of commuting conditions using mobile signaling data,but there is still limited research on travel mode recognition based on mobile signaling data.In this context,on the basis of in-depth mining of residents’ travel characteristics,this paper proposes a travel mode recognition model based on residents’ travel characteristics data.Combined with three machine learning algorithms,namely support vector machine,BP neural network,and Random forest,this paper identifies five common travel modes of urban residents,namely,walking,cycling,car travel,bus travel,and subway travel,for data with and without travel mode labels.First of all,the original processing and Data cleansing of mobile signaling data are carried out,and the extraction of residents’ travel chain and map matching of mobile signaling data are carried out,laying the foundation for the extraction and processing of travel characteristics.Secondly,combined with urban geographic space and resident travel survey data,obtain travel mode labels for resident travel chain data extracted from mobile signaling data.In terms of feature data extraction,features such as resident travel time,travel space,travel distance,travel speed,travel purpose,and various social attributes of travelers were obtained for travel mode recognition.Statistical analysis was conducted on resident travel feature data to ensure that the feature data has an impact on the recognition of travel modes;In terms of feature data processing,the missing data,duplicate values and Outlier are cleaned,and then data normalization is performed to ensure the equivalence between the model construction data set and the model application data set,thus ensuring the validity of the model results.Thirdly,select model strategies by analyzing the principles of three machine learning algorithms.Taking the influence of parameters on accuracy as the criterion,the support vector machine,BP neural network and Random forest machine learning models are established on the premise of establishing the parameter optimization scheme and setting the parameters of the three models.Finally,a case study of travel mode identification for Harbin residents is presented.In this study,the results of model recognition were analyzed in detail,and the accuracy rates of the three models were 65.25%,71.25%,and 75.35%,respectively.Comprehensive comparison and evaluation were conducted by using multiple indicators such as Learning curve,accuracy,precision,recall,F1 score,AUC value,and Receiver operating characteristic.The evaluation results show that the performance of the model meets the expectations and has a good application prospect,and the Random forest model has the best classification and recognition effect.Based on the best model results,importance analysis was conducted on various features,with distance,average speed,and age being the three most important features,accounting for18.36%,17.51%,and 10.67% of the total importance,respectively.The unlabeled travel chain was identified for travel modes,indicating that the prediction results of the resident travel feature data classification recognition model are stable and reliable,and have certain application value.