Extracting Metro Passenger Flow Predictors from Network’s Complex Characteristics

Complex network characteristics such as centralities have lately started to be associated with passenger flows at metro stations. Centralities can be leveraged in an effort to develop fast and cost-efficient passenger flow predictive models. However, the accuracy of such models is still under question and the most appropriate predictors are yet to be found. In this sense, this study attempts to investigate appropriate predictors, and develop a predictive model for daily passenger flows at metro stations, based exclusively on spatial attributes. Using the Athens metro network as a case study, a linear regression model is developed, with node degree, betweenness and closeness centralities of the physical network, node strength of the substitute network, and a dummy variable of station importance being as covariates. An econometric analysis validates that a linear model is suitable for associating centralities with passenger flows, while model’s evaluation metrics indicate satisfying accuracy. In addition, a machine learning benchmark model is utilized to further investigate variable significance and validate the accuracy of the linear model. Last but not least, both models are utilized for predicting passenger flows at the new metro stations of the Athens metro network expansion. Findings suggest that node strength of the substitute network is a powerful predictor and the most significant covariate of both models; both models’ accuracy and predictions converge to a great extent. The model developed is expected to facilitate medium-term disruption management through providing information about metro passenger flows at low cost and high speed.