An Age Attribute Inference Method for Urban Rail Transit Passengers based on Deep Learning

The age information of urban rail transit passengers is of great significance for accurate passenger flow organization and personalized intelligent service. Aiming at the problems of missing and difficult to obtain the passenger age attribute of rail transit, this paper integrates multi-source data mining such as automatic fare collection (AFC) data and urban land use data to reflect the travel characteristics of passenger age, and proposes a passenger age attribute inference model combining deep neural network (DNN) and automatic encoder (AE). Guangzhou Metro is selected for case analysis. The results show that compared with the baseline model SVM, MLP, DT and AdaBoost, the accuracy of DNN+AE model proposed in this paper is improved by 11.08%, 9.82%, 5.33% and 3.33%, respectively. Among them, the accuracy of age attribute inference for the elderly is the highest, reaching 78.58%.